Made multigpu deepclone load model from disk to avoid needing to deepclone actual model object, fixed issues with merge, turn off cuda backend as it causes device mismatch issue with rope (and potentially other ops), will investigate

.coderabbit.yaml

@@ -1,114 +0,0 @@
-# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
-language: "en-US"
-early_access: false
-
-reviews:
-  profile: "chill"
-  request_changes_workflow: false
-  high_level_summary: false
-  poem: false
-  review_status: false
-  review_details: false
-  commit_status: true
-  collapse_walkthrough: true
-  changed_files_summary: false
-  sequence_diagrams: false
-  estimate_code_review_effort: false
-  assess_linked_issues: false
-  related_issues: false
-  related_prs: false
-  suggested_labels: false
-  auto_apply_labels: false
-  suggested_reviewers: false
-  auto_assign_reviewers: false
-  in_progress_fortune: false
-  enable_prompt_for_ai_agents: true
-
-  path_filters:
-    - "!comfy_api_nodes/apis/**"
-    - "!**/generated/*.pyi"
-    - "!.ci/**"
-    - "!script_examples/**"
-    - "!**/__pycache__/**"
-    - "!**/*.ipynb"
-    - "!**/*.png"
-    - "!**/*.bat"
-
-  path_instructions:
-    - path: "comfy/**"
-      instructions: |
-        Core ML/diffusion engine. Focus on:
-        - Backward compatibility (breaking changes affect all custom nodes)
-        - Memory management and GPU resource handling
-        - Performance implications in hot paths
-        - Thread safety for concurrent execution
-    - path: "comfy_api_nodes/**"
-      instructions: |
-        Third-party API integration nodes. Focus on:
-        - No hardcoded API keys or secrets
-        - Proper error handling for API failures (timeouts, rate limits, auth errors)
-        - Correct Pydantic model usage
-        - Security of user data passed to external APIs
-    - path: "comfy_extras/**"
-      instructions: |
-        Community-contributed extra nodes. Focus on:
-        - Consistency with node patterns (INPUT_TYPES, RETURN_TYPES, FUNCTION, CATEGORY)
-        - No breaking changes to existing node interfaces
-    - path: "comfy_execution/**"
-      instructions: |
-        Execution engine (graph execution, caching, jobs). Focus on:
-        - Caching correctness
-        - Concurrent execution safety
-        - Graph validation edge cases
-    - path: "nodes.py"
-      instructions: |
-        Core node definitions (2500+ lines). Focus on:
-        - Backward compatibility of NODE_CLASS_MAPPINGS
-        - Consistency of INPUT_TYPES return format
-    - path: "alembic_db/**"
-      instructions: |
-        Database migrations. Focus on:
-        - Migration safety and rollback support
-        - Data preservation during schema changes
-
-  auto_review:
-    enabled: true
-    auto_incremental_review: true
-    drafts: true
-
-  finishing_touches:
-    docstrings:
-      enabled: false
-    unit_tests:
-      enabled: false
-
-  tools:
-    ruff:
-      enabled: true
-    pylint:
-      enabled: false
-    flake8:
-      enabled: false
-    gitleaks:
-      enabled: true
-    shellcheck:
-      enabled: false
-    markdownlint:
-      enabled: false
-    yamllint:
-      enabled: false
-    languagetool:
-      enabled: false
-    github-checks:
-      enabled: true
-      timeout_ms: 90000
-    ast-grep:
-      essential_rules: true
-
-chat:
-  auto_reply: true
-
-knowledge_base:
-  opt_out: false
-  learnings:
-    scope: "auto"

README.md

@@ -229,9 +229,9 @@ AMD users can install rocm and pytorch with pip if you don't have it already ins
 
 ```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.1```
 
-This is the command to install the nightly with ROCm 7.2 which might have some performance improvements:
+This is the command to install the nightly with ROCm 7.1 which might have some performance improvements:
 
-```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.2```
+```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.1```
 
 
 ### AMD GPUs (Experimental: Windows and Linux), RDNA 3, 3.5 and 4 only.

blueprints/.glsl/Brightness_and_Contrast_1.frag

@@ -1,44 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0; // Brightness slider -100..100
-uniform float u_float1; // Contrast slider -100..100
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const float MID_GRAY = 0.18;  // 18% reflectance
-
-// sRGB gamma 2.2 approximation
-vec3 srgbToLinear(vec3 c) {
-    return pow(max(c, 0.0), vec3(2.2));
-}
-
-vec3 linearToSrgb(vec3 c) {
-    return pow(max(c, 0.0), vec3(1.0/2.2));
-}
-
-float mapBrightness(float b) {
-    return clamp(b / 100.0, -1.0, 1.0);
-}
-
-float mapContrast(float c) {
-    return clamp(c / 100.0 + 1.0, 0.0, 2.0);
-}
-
-void main() {
-    vec4 orig = texture(u_image0, v_texCoord);
-
-    float brightness = mapBrightness(u_float0);
-    float contrast   = mapContrast(u_float1);
-
-    vec3 lin = srgbToLinear(orig.rgb);
-
-    lin = (lin - MID_GRAY) * contrast + brightness + MID_GRAY;
-
-    // Convert back to sRGB
-    vec3 result = linearToSrgb(clamp(lin, 0.0, 1.0));
-
-    fragColor = vec4(result, orig.a);
-}

blueprints/.glsl/Chromatic_Aberration_16.frag

@@ -1,72 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Mode
-uniform float u_float0;  // Amount (0 to 100)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int MODE_LINEAR   = 0;
-const int MODE_RADIAL   = 1;
-const int MODE_BARREL   = 2;
-const int MODE_SWIRL    = 3;
-const int MODE_DIAGONAL = 4;
-
-const float AMOUNT_SCALE = 0.0005;
-const float RADIAL_MULT = 4.0;
-const float BARREL_MULT = 8.0;
-const float INV_SQRT2 = 0.70710678118;
-
-void main() {
-    vec2 uv = v_texCoord;
-    vec4 original = texture(u_image0, uv);
-
-    float amount = u_float0 * AMOUNT_SCALE;
-
-    if (amount < 0.000001) {
-        fragColor = original;
-        return;
-    }
-
-    // Aspect-corrected coordinates for circular effects
-    float aspect = u_resolution.x / u_resolution.y;
-    vec2 centered = uv - 0.5;
-    vec2 corrected = vec2(centered.x * aspect, centered.y);
-    float r = length(corrected);
-    vec2 dir = r > 0.0001 ? corrected / r : vec2(0.0);
-    vec2 offset = vec2(0.0);
-
-    if (u_int0 == MODE_LINEAR) {
-        // Horizontal shift (no aspect correction needed)
-        offset = vec2(amount, 0.0);
-    }
-    else if (u_int0 == MODE_RADIAL) {
-        // Outward from center, stronger at edges
-        offset = dir * r * amount * RADIAL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_BARREL) {
-        // Lens distortion simulation (r² falloff)
-        offset = dir * r * r * amount * BARREL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_SWIRL) {
-        // Perpendicular to radial (rotational aberration)
-        vec2 perp = vec2(-dir.y, dir.x);
-        offset = perp * r * amount * RADIAL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_DIAGONAL) {
-        // 45° offset (no aspect correction needed)
-        offset = vec2(amount, amount) * INV_SQRT2;
-    }
-    
-    float red = texture(u_image0, uv + offset).r;
-    float green = original.g;
-    float blue = texture(u_image0, uv - offset).b;
-    
-    fragColor = vec4(red, green, blue, original.a);
-}

blueprints/.glsl/Color_Adjustment_15.frag

@@ -1,78 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0; // temperature (-100 to 100)
-uniform float u_float1; // tint (-100 to 100)
-uniform float u_float2; // vibrance (-100 to 100)
-uniform float u_float3; // saturation (-100 to 100)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const float INPUT_SCALE = 0.01;
-const float TEMP_TINT_PRIMARY = 0.3;
-const float TEMP_TINT_SECONDARY = 0.15;
-const float VIBRANCE_BOOST = 2.0;
-const float SATURATION_BOOST = 2.0;
-const float SKIN_PROTECTION = 0.5;
-const float EPSILON = 0.001;
-const vec3 LUMA_WEIGHTS = vec3(0.299, 0.587, 0.114);
-
-void main() {
-    vec4 tex = texture(u_image0, v_texCoord);
-    vec3 color = tex.rgb;
-    
-    // Scale inputs: -100/100 → -1/1
-    float temperature = u_float0 * INPUT_SCALE;
-    float tint = u_float1 * INPUT_SCALE;
-    float vibrance = u_float2 * INPUT_SCALE;
-    float saturation = u_float3 * INPUT_SCALE;
-    
-    // Temperature (warm/cool): positive = warm, negative = cool
-    color.r += temperature * TEMP_TINT_PRIMARY;
-    color.b -= temperature * TEMP_TINT_PRIMARY;
-    
-    // Tint (green/magenta): positive = green, negative = magenta
-    color.g += tint * TEMP_TINT_PRIMARY;
-    color.r -= tint * TEMP_TINT_SECONDARY;
-    color.b -= tint * TEMP_TINT_SECONDARY;
-    
-    // Single clamp after temperature/tint
-    color = clamp(color, 0.0, 1.0);
-    
-    // Vibrance with skin protection
-    if (vibrance != 0.0) {
-        float maxC = max(color.r, max(color.g, color.b));
-        float minC = min(color.r, min(color.g, color.b));
-        float sat = maxC - minC;
-        float gray = dot(color, LUMA_WEIGHTS);
-        
-        if (vibrance < 0.0) {
-            // Desaturate: -100 → gray
-            color = mix(vec3(gray), color, 1.0 + vibrance);
-        } else {
-            // Boost less saturated colors more
-            float vibranceAmt = vibrance * (1.0 - sat);
-            
-            // Branchless skin tone protection
-            float isWarmTone = step(color.b, color.g) * step(color.g, color.r);
-            float warmth = (color.r - color.b) / max(maxC, EPSILON);
-            float skinTone = isWarmTone * warmth * sat * (1.0 - sat);
-            vibranceAmt *= (1.0 - skinTone * SKIN_PROTECTION);
-            
-            color = mix(vec3(gray), color, 1.0 + vibranceAmt * VIBRANCE_BOOST);
-        }
-    }
-    
-    // Saturation
-    if (saturation != 0.0) {
-        float gray = dot(color, LUMA_WEIGHTS);
-        float satMix = saturation < 0.0
-            ? 1.0 + saturation                      // -100 → gray
-            : 1.0 + saturation * SATURATION_BOOST;  // +100 → 3x boost
-        color = mix(vec3(gray), color, satMix);
-    }
-    
-    fragColor = vec4(clamp(color, 0.0, 1.0), tex.a);
-}

blueprints/.glsl/Edge-Preserving_Blur_128.frag

@@ -1,94 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0;   // Blur radius (0–20, default ~5)
-uniform float u_float1;   // Edge threshold (0–100, default ~30)
-uniform int u_int0;       // Step size (0/1 = every pixel, 2+ = skip pixels)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int MAX_RADIUS = 20;
-const float EPSILON = 0.0001;
-
-// Perceptual luminance
-float getLuminance(vec3 rgb) {
-    return dot(rgb, vec3(0.299, 0.587, 0.114));
-}
-
-vec4 bilateralFilter(vec2 uv, vec2 texelSize, int radius,
-                     float sigmaSpatial, float sigmaColor)
-{
-    vec4 center = texture(u_image0, uv);
-    vec3 centerRGB = center.rgb;
-
-    float invSpatial2 = -0.5 / (sigmaSpatial * sigmaSpatial);
-    float invColor2   = -0.5 / (sigmaColor * sigmaColor + EPSILON);
-
-    vec3 sumRGB = vec3(0.0);
-    float sumWeight = 0.0;
-
-    int step = max(u_int0, 1);
-    float radius2 = float(radius * radius);
-
-    for (int dy = -MAX_RADIUS; dy <= MAX_RADIUS; dy++) {
-        if (dy < -radius || dy > radius) continue;
-        if (abs(dy) % step != 0) continue;
-
-        for (int dx = -MAX_RADIUS; dx <= MAX_RADIUS; dx++) {
-            if (dx < -radius || dx > radius) continue;
-            if (abs(dx) % step != 0) continue;
-
-            vec2 offset = vec2(float(dx), float(dy));
-            float dist2 = dot(offset, offset);
-            if (dist2 > radius2) continue;
-
-            vec3 sampleRGB = texture(u_image0, uv + offset * texelSize).rgb;
-
-            // Spatial Gaussian
-            float spatialWeight = exp(dist2 * invSpatial2);
-
-            // Perceptual color distance (weighted RGB)
-            vec3 diff = sampleRGB - centerRGB;
-            float colorDist = dot(diff * diff, vec3(0.299, 0.587, 0.114));
-            float colorWeight = exp(colorDist * invColor2);
-
-            float w = spatialWeight * colorWeight;
-            sumRGB += sampleRGB * w;
-            sumWeight += w;
-        }
-    }
-
-    vec3 resultRGB = sumRGB / max(sumWeight, EPSILON);
-    return vec4(resultRGB, center.a); // preserve center alpha
-}
-
-void main() {
-    vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));
-
-    float radiusF = clamp(u_float0, 0.0, float(MAX_RADIUS));
-    int radius = int(radiusF + 0.5);
-
-    if (radius == 0) {
-        fragColor = texture(u_image0, v_texCoord);
-        return;
-    }
-
-    // Edge threshold → color sigma
-    // Squared curve for better low-end control
-    float t = clamp(u_float1, 0.0, 100.0) / 100.0;
-    t *= t;
-    float sigmaColor = mix(0.01, 0.5, t);
-
-    // Spatial sigma tied to radius
-    float sigmaSpatial = max(radiusF * 0.75, 0.5);
-
-    fragColor = bilateralFilter(
-        v_texCoord,
-        texelSize,
-        radius,
-        sigmaSpatial,
-        sigmaColor
-    );
-}

blueprints/.glsl/Film_Grain_15.frag

@@ -1,124 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0; // grain amount      [0.0 – 1.0]   typical: 0.2–0.8
-uniform float u_float1; // grain size        [0.3 – 3.0]   lower = finer grain
-uniform float u_float2; // color amount      [0.0 – 1.0]   0 = monochrome, 1 = RGB grain
-uniform float u_float3; // luminance bias    [0.0 – 1.0]   0 = uniform, 1 = shadows only
-uniform int   u_int0;   // noise mode        [0 or 1]      0 = smooth, 1 = grainy
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-// High-quality integer hash (pcg-like)
-uint pcg(uint v) {
-    uint state = v * 747796405u + 2891336453u;
-    uint word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;
-    return (word >> 22u) ^ word;
-}
-
-// 2D -> 1D hash input
-uint hash2d(uvec2 p) {
-    return pcg(p.x + pcg(p.y));
-}
-
-// Hash to float [0, 1]
-float hashf(uvec2 p) {
-    return float(hash2d(p)) / float(0xffffffffu);
-}
-
-// Hash to float with offset (for RGB channels)
-float hashf(uvec2 p, uint offset) {
-    return float(pcg(hash2d(p) + offset)) / float(0xffffffffu);
-}
-
-// Convert uniform [0,1] to roughly Gaussian distribution
-// Using simple approximation: average of multiple samples
-float toGaussian(uvec2 p) {
-    float sum = hashf(p, 0u) + hashf(p, 1u) + hashf(p, 2u) + hashf(p, 3u);
-    return (sum - 2.0) * 0.7;  // Centered, scaled
-}
-
-float toGaussian(uvec2 p, uint offset) {
-    float sum = hashf(p, offset) + hashf(p, offset + 1u) 
-              + hashf(p, offset + 2u) + hashf(p, offset + 3u);
-    return (sum - 2.0) * 0.7;
-}
-
-// Smooth noise with better interpolation
-float smoothNoise(vec2 p) {
-    vec2 i = floor(p);
-    vec2 f = fract(p);
-    
-    // Quintic interpolation (less banding than cubic)
-    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);
-    
-    uvec2 ui = uvec2(i);
-    float a = toGaussian(ui);
-    float b = toGaussian(ui + uvec2(1u, 0u));
-    float c = toGaussian(ui + uvec2(0u, 1u));
-    float d = toGaussian(ui + uvec2(1u, 1u));
-    
-    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);
-}
-
-float smoothNoise(vec2 p, uint offset) {
-    vec2 i = floor(p);
-    vec2 f = fract(p);
-    
-    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);
-    
-    uvec2 ui = uvec2(i);
-    float a = toGaussian(ui, offset);
-    float b = toGaussian(ui + uvec2(1u, 0u), offset);
-    float c = toGaussian(ui + uvec2(0u, 1u), offset);
-    float d = toGaussian(ui + uvec2(1u, 1u), offset);
-    
-    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);
-}
-
-void main() {
-    vec4 color = texture(u_image0, v_texCoord);
-    
-    // Luminance (Rec.709)
-    float luma = dot(color.rgb, vec3(0.2126, 0.7152, 0.0722));
-    
-    // Grain UV (resolution-independent)
-    vec2 grainUV = v_texCoord * u_resolution / max(u_float1, 0.01);
-    uvec2 grainPixel = uvec2(grainUV);
-    
-    float g;
-    vec3 grainRGB;
-    
-    if (u_int0 == 1) {
-        // Grainy mode: pure hash noise (no interpolation = no banding)
-        g = toGaussian(grainPixel);
-        grainRGB = vec3(
-            toGaussian(grainPixel, 100u),
-            toGaussian(grainPixel, 200u),
-            toGaussian(grainPixel, 300u)
-        );
-    } else {
-        // Smooth mode: interpolated with quintic curve
-        g = smoothNoise(grainUV);
-        grainRGB = vec3(
-            smoothNoise(grainUV, 100u),
-            smoothNoise(grainUV, 200u),
-            smoothNoise(grainUV, 300u)
-        );
-    }
-    
-    // Luminance weighting (less grain in highlights)
-    float lumWeight = mix(1.0, 1.0 - luma, clamp(u_float3, 0.0, 1.0));
-    
-    // Strength
-    float strength = u_float0 * 0.15;
-    
-    // Color vs monochrome grain
-    vec3 grainColor = mix(vec3(g), grainRGB, clamp(u_float2, 0.0, 1.0));
-    
-    color.rgb += grainColor * strength * lumWeight;
-    fragColor0 = vec4(clamp(color.rgb, 0.0, 1.0), color.a);
-}

blueprints/.glsl/Glow_30.frag

@@ -1,133 +0,0 @@
-#version 300 es
-precision mediump float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Blend mode
-uniform int u_int1;      // Color tint
-uniform float u_float0;  // Intensity
-uniform float u_float1;  // Radius
-uniform float u_float2;  // Threshold
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int BLEND_ADD      = 0;
-const int BLEND_SCREEN   = 1;
-const int BLEND_SOFT     = 2;
-const int BLEND_OVERLAY  = 3;
-const int BLEND_LIGHTEN  = 4;
-
-const float GOLDEN_ANGLE = 2.39996323;
-const int MAX_SAMPLES = 48;
-const vec3 LUMA = vec3(0.299, 0.587, 0.114);
-
-float hash(vec2 p) {
-    p = fract(p * vec2(123.34, 456.21));
-    p += dot(p, p + 45.32);
-    return fract(p.x * p.y);
-}
-
-vec3 hexToRgb(int h) {
-    return vec3(
-        float((h >> 16) & 255),
-        float((h >> 8) & 255),
-        float(h & 255)
-    ) * (1.0 / 255.0);
-}
-
-vec3 blend(vec3 base, vec3 glow, int mode) {
-    if (mode == BLEND_SCREEN) {
-        return 1.0 - (1.0 - base) * (1.0 - glow);
-    }
-    if (mode == BLEND_SOFT) {
-        return mix(
-            base - (1.0 - 2.0 * glow) * base * (1.0 - base),
-            base + (2.0 * glow - 1.0) * (sqrt(base) - base),
-            step(0.5, glow)
-        );
-    }
-    if (mode == BLEND_OVERLAY) {
-        return mix(
-            2.0 * base * glow,
-            1.0 - 2.0 * (1.0 - base) * (1.0 - glow),
-            step(0.5, base)
-        );
-    }
-    if (mode == BLEND_LIGHTEN) {
-        return max(base, glow);
-    }
-    return base + glow;
-}
-
-void main() {
-    vec4 original = texture(u_image0, v_texCoord);
-    
-    float intensity = u_float0 * 0.05;
-    float radius = u_float1 * u_float1 * 0.012;
-    
-    if (intensity < 0.001 || radius < 0.1) {
-        fragColor = original;
-        return;
-    }
-    
-    float threshold = 1.0 - u_float2 * 0.01;
-    float t0 = threshold - 0.15;
-    float t1 = threshold + 0.15;
-    
-    vec2 texelSize = 1.0 / u_resolution;
-    float radius2 = radius * radius;
-    
-    float sampleScale = clamp(radius * 0.75, 0.35, 1.0);
-    int samples = int(float(MAX_SAMPLES) * sampleScale);
-    
-    float noise = hash(gl_FragCoord.xy);
-    float angleOffset = noise * GOLDEN_ANGLE;
-    float radiusJitter = 0.85 + noise * 0.3;
-    
-    float ca = cos(GOLDEN_ANGLE);
-    float sa = sin(GOLDEN_ANGLE);
-    vec2 dir = vec2(cos(angleOffset), sin(angleOffset));
-    
-    vec3 glow = vec3(0.0);
-    float totalWeight = 0.0;
-    
-    // Center tap
-    float centerMask = smoothstep(t0, t1, dot(original.rgb, LUMA));
-    glow += original.rgb * centerMask * 2.0;
-    totalWeight += 2.0;
-    
-    for (int i = 1; i < MAX_SAMPLES; i++) {
-        if (i >= samples) break;
-        
-        float fi = float(i);
-        float dist = sqrt(fi / float(samples)) * radius * radiusJitter;
-        
-        vec2 offset = dir * dist * texelSize;
-        vec3 c = texture(u_image0, v_texCoord + offset).rgb;
-        float mask = smoothstep(t0, t1, dot(c, LUMA));
-        
-        float w = 1.0 - (dist * dist) / (radius2 * 1.5);
-        w = max(w, 0.0);
-        w *= w;
-        
-        glow += c * mask * w;
-        totalWeight += w;
-        
-        dir = vec2(
-            dir.x * ca - dir.y * sa,
-            dir.x * sa + dir.y * ca
-        );
-    }
-    
-    glow *= intensity / max(totalWeight, 0.001);
-    
-    if (u_int1 > 0) {
-        glow *= hexToRgb(u_int1);
-    }
-    
-    vec3 result = blend(original.rgb, glow, u_int0);
-    result += (noise - 0.5) * (1.0 / 255.0);
-    
-    fragColor = vec4(clamp(result, 0.0, 1.0), original.a);
-}

blueprints/.glsl/Hue_and_Saturation_1.frag

@@ -1,222 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform int u_int0;      // Mode: 0=Master, 1=Reds, 2=Yellows, 3=Greens, 4=Cyans, 5=Blues, 6=Magentas, 7=Colorize
-uniform int u_int1;      // Color Space: 0=HSL, 1=HSB/HSV
-uniform float u_float0;  // Hue (-180 to 180)
-uniform float u_float1;  // Saturation (-100 to 100)
-uniform float u_float2;  // Lightness/Brightness (-100 to 100)
-uniform float u_float3;  // Overlap (0 to 100) - feathering between adjacent color ranges
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-// Color range modes
-const int MODE_MASTER   = 0;
-const int MODE_RED      = 1;
-const int MODE_YELLOW   = 2;
-const int MODE_GREEN    = 3;
-const int MODE_CYAN     = 4;
-const int MODE_BLUE     = 5;
-const int MODE_MAGENTA  = 6;
-const int MODE_COLORIZE = 7;
-
-// Color space modes
-const int COLORSPACE_HSL = 0;
-const int COLORSPACE_HSB = 1;
-
-const float EPSILON = 0.0001;
-
-//=============================================================================
-// RGB <-> HSL Conversions
-//=============================================================================
-
-vec3 rgb2hsl(vec3 c) {
-    float maxC = max(max(c.r, c.g), c.b);
-    float minC = min(min(c.r, c.g), c.b);
-    float delta = maxC - minC;
-
-    float h = 0.0;
-    float s = 0.0;
-    float l = (maxC + minC) * 0.5;
-
-    if (delta > EPSILON) {
-        s = l < 0.5
-            ? delta / (maxC + minC)
-            : delta / (2.0 - maxC - minC);
-
-        if (maxC == c.r) {
-            h = (c.g - c.b) / delta + (c.g < c.b ? 6.0 : 0.0);
-        } else if (maxC == c.g) {
-            h = (c.b - c.r) / delta + 2.0;
-        } else {
-            h = (c.r - c.g) / delta + 4.0;
-        }
-        h /= 6.0;
-    }
-
-    return vec3(h, s, l);
-}
-
-float hue2rgb(float p, float q, float t) {
-    t = fract(t);
-    if (t < 1.0/6.0) return p + (q - p) * 6.0 * t;
-    if (t < 0.5)       return q;
-    if (t < 2.0/3.0)   return p + (q - p) * (2.0/3.0 - t) * 6.0;
-    return p;
-}
-
-vec3 hsl2rgb(vec3 hsl) {
-    if (hsl.y < EPSILON) return vec3(hsl.z);
-
-    float q = hsl.z < 0.5
-        ? hsl.z * (1.0 + hsl.y)
-        : hsl.z + hsl.y - hsl.z * hsl.y;
-    float p = 2.0 * hsl.z - q;
-
-    return vec3(
-        hue2rgb(p, q, hsl.x + 1.0/3.0),
-        hue2rgb(p, q, hsl.x),
-        hue2rgb(p, q, hsl.x - 1.0/3.0)
-    );
-}
-
-vec3 rgb2hsb(vec3 c) {
-    float maxC = max(max(c.r, c.g), c.b);
-    float minC = min(min(c.r, c.g), c.b);
-    float delta = maxC - minC;
-
-    float h = 0.0;
-    float s = (maxC > EPSILON) ? delta / maxC : 0.0;
-    float b = maxC;
-
-    if (delta > EPSILON) {
-        if (maxC == c.r) {
-            h = (c.g - c.b) / delta + (c.g < c.b ? 6.0 : 0.0);
-        } else if (maxC == c.g) {
-            h = (c.b - c.r) / delta + 2.0;
-        } else {
-            h = (c.r - c.g) / delta + 4.0;
-        }
-        h /= 6.0;
-    }
-
-    return vec3(h, s, b);
-}
-
-vec3 hsb2rgb(vec3 hsb) {
-    vec3 rgb = clamp(abs(mod(hsb.x * 6.0 + vec3(0.0, 4.0, 2.0), 6.0) - 3.0) - 1.0, 0.0, 1.0);
-    return hsb.z * mix(vec3(1.0), rgb, hsb.y);
-}
-
-//=============================================================================
-// Color Range Weight Calculation
-//=============================================================================
-
-float hueDistance(float a, float b) {
-    float d = abs(a - b);
-    return min(d, 1.0 - d);
-}
-
-float getHueWeight(float hue, float center, float overlap) {
-    float baseWidth = 1.0 / 6.0;
-    float feather = baseWidth * overlap;
-
-    float d = hueDistance(hue, center);
-
-    float inner = baseWidth * 0.5;
-    float outer = inner + feather;
-
-    return 1.0 - smoothstep(inner, outer, d);
-}
-
-float getModeWeight(float hue, int mode, float overlap) {
-    if (mode == MODE_MASTER || mode == MODE_COLORIZE) return 1.0;
-
-    if (mode == MODE_RED) {
-        return max(
-            getHueWeight(hue, 0.0, overlap),
-            getHueWeight(hue, 1.0, overlap)
-        );
-    }
-
-    float center = float(mode - 1) / 6.0;
-    return getHueWeight(hue, center, overlap);
-}
-
-//=============================================================================
-// Adjustment Functions
-//=============================================================================
-
-float adjustLightness(float l, float amount) {
-    return amount > 0.0
-        ? l + (1.0 - l) * amount
-        : l + l * amount;
-}
-
-float adjustBrightness(float b, float amount) {
-    return clamp(b + amount, 0.0, 1.0);
-}
-
-float adjustSaturation(float s, float amount) {
-    return amount > 0.0
-        ? s + (1.0 - s) * amount
-        : s + s * amount;
-}
-
-vec3 colorize(vec3 rgb, float hue, float sat, float light) {
-    float lum = dot(rgb, vec3(0.299, 0.587, 0.114));
-    float l = adjustLightness(lum, light);
-
-    vec3 hsl = vec3(fract(hue), clamp(sat, 0.0, 1.0), clamp(l, 0.0, 1.0));
-    return hsl2rgb(hsl);
-}
-
-//=============================================================================
-// Main
-//=============================================================================
-
-void main() {
-    vec4 original = texture(u_image0, v_texCoord);
-
-    float hueShift   = u_float0 / 360.0;   // -180..180 -> -0.5..0.5
-    float satAmount  = u_float1 / 100.0;   // -100..100 -> -1..1
-    float lightAmount= u_float2 / 100.0;   // -100..100 -> -1..1
-    float overlap    = u_float3 / 100.0;   // 0..100 -> 0..1
-
-    vec3 result;
-
-    if (u_int0 == MODE_COLORIZE) {
-        result = colorize(original.rgb, hueShift, satAmount, lightAmount);
-        fragColor = vec4(result, original.a);
-        return;
-    }
-
-    vec3 hsx = (u_int1 == COLORSPACE_HSL)
-        ? rgb2hsl(original.rgb)
-        : rgb2hsb(original.rgb);
-
-    float weight = getModeWeight(hsx.x, u_int0, overlap);
-
-    if (u_int0 != MODE_MASTER && hsx.y < EPSILON) {
-        weight = 0.0;
-    }
-
-    if (weight > EPSILON) {
-        float h = fract(hsx.x + hueShift * weight);
-        float s = clamp(adjustSaturation(hsx.y, satAmount * weight), 0.0, 1.0);
-        float v = (u_int1 == COLORSPACE_HSL)
-            ? clamp(adjustLightness(hsx.z, lightAmount * weight), 0.0, 1.0)
-            : clamp(adjustBrightness(hsx.z, lightAmount * weight), 0.0, 1.0);
-
-        vec3 adjusted = vec3(h, s, v);
-        result = (u_int1 == COLORSPACE_HSL)
-            ? hsl2rgb(adjusted)
-            : hsb2rgb(adjusted);
-    } else {
-        result = original.rgb;
-    }
-
-    fragColor = vec4(result, original.a);
-}

blueprints/.glsl/Image_Blur_1.frag

@@ -1,111 +0,0 @@
-#version 300 es
-#pragma passes 2
-precision highp float;
-
-// Blur type constants
-const int BLUR_GAUSSIAN = 0;
-const int BLUR_BOX = 1;
-const int BLUR_RADIAL = 2;
-
-// Radial blur config
-const int RADIAL_SAMPLES = 12;
-const float RADIAL_STRENGTH = 0.0003;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL)
-uniform float u_float0;  // Blur radius/amount
-uniform int u_pass;      // Pass index (0 = horizontal, 1 = vertical)
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-float gaussian(float x, float sigma) {
-    return exp(-(x * x) / (2.0 * sigma * sigma));
-}
-
-void main() {
-    vec2 texelSize = 1.0 / u_resolution;
-    float radius = max(u_float0, 0.0);
-
-    // Radial (angular) blur - single pass, doesn't use separable
-    if (u_int0 == BLUR_RADIAL) {
-        // Only execute on first pass
-        if (u_pass > 0) {
-            fragColor0 = texture(u_image0, v_texCoord);
-            return;
-        }
-
-        vec2 center = vec2(0.5);
-        vec2 dir = v_texCoord - center;
-        float dist = length(dir);
-
-        if (dist < 1e-4) {
-            fragColor0 = texture(u_image0, v_texCoord);
-            return;
-        }
-
-        vec4 sum = vec4(0.0);
-        float totalWeight = 0.0;
-        float angleStep = radius * RADIAL_STRENGTH;
-
-        dir /= dist;
-
-        float cosStep = cos(angleStep);
-        float sinStep = sin(angleStep);
-
-        float negAngle = -float(RADIAL_SAMPLES) * angleStep;
-        vec2 rotDir = vec2(
-            dir.x * cos(negAngle) - dir.y * sin(negAngle),
-            dir.x * sin(negAngle) + dir.y * cos(negAngle)
-        );
-
-        for (int i = -RADIAL_SAMPLES; i <= RADIAL_SAMPLES; i++) {
-            vec2 uv = center + rotDir * dist;
-            float w = 1.0 - abs(float(i)) / float(RADIAL_SAMPLES);
-            sum += texture(u_image0, uv) * w;
-            totalWeight += w;
-
-            rotDir = vec2(
-                rotDir.x * cosStep - rotDir.y * sinStep,
-                rotDir.x * sinStep + rotDir.y * cosStep
-            );
-        }
-
-        fragColor0 = sum / max(totalWeight, 0.001);
-        return;
-    }
-
-    // Separable Gaussian / Box blur
-    int samples = int(ceil(radius));
-
-    if (samples == 0) {
-        fragColor0 = texture(u_image0, v_texCoord);
-        return;
-    }
-
-    // Direction: pass 0 = horizontal, pass 1 = vertical
-    vec2 dir = (u_pass == 0) ? vec2(1.0, 0.0) : vec2(0.0, 1.0);
-
-    vec4 color = vec4(0.0);
-    float totalWeight = 0.0;
-    float sigma = radius / 2.0;
-
-    for (int i = -samples; i <= samples; i++) {
-        vec2 offset = dir * float(i) * texelSize;
-        vec4 sample_color = texture(u_image0, v_texCoord + offset);
-
-        float weight;
-        if (u_int0 == BLUR_GAUSSIAN) {
-            weight = gaussian(float(i), sigma);
-        } else {
-            // BLUR_BOX
-            weight = 1.0;
-        }
-
-        color += sample_color * weight;
-        totalWeight += weight;
-    }
-
-    fragColor0 = color / totalWeight;
-}

blueprints/.glsl/Image_Channels_23.frag

@@ -1,19 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-layout(location = 1) out vec4 fragColor1;
-layout(location = 2) out vec4 fragColor2;
-layout(location = 3) out vec4 fragColor3;
-
-void main() {
-  vec4 color = texture(u_image0, v_texCoord);
-  // Output each channel as grayscale to separate render targets
-  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel
-  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel
-  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel
-  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel
-}

blueprints/.glsl/Image_Levels_1.frag

@@ -1,71 +0,0 @@
-#version 300 es
-precision highp float;
-
-// Levels Adjustment
-// u_int0:   channel      (0=RGB, 1=R, 2=G, 3=B)         default: 0
-// u_float0: input black  (0-255)                        default: 0
-// u_float1: input white  (0-255)                        default: 255
-// u_float2: gamma        (0.01-9.99)                    default: 1.0
-// u_float3: output black (0-255)                        default: 0
-// u_float4: output white (0-255)                        default: 255
-
-uniform sampler2D u_image0;
-uniform int u_int0;
-uniform float u_float0;
-uniform float u_float1;
-uniform float u_float2;
-uniform float u_float3;
-uniform float u_float4;
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-vec3 applyLevels(vec3 color, float inBlack, float inWhite, float gamma, float outBlack, float outWhite) {
-    float inRange = max(inWhite - inBlack, 0.0001);
-    vec3 result = clamp((color - inBlack) / inRange, 0.0, 1.0);
-    result = pow(result, vec3(1.0 / gamma));
-    result = mix(vec3(outBlack), vec3(outWhite), result);
-    return result;
-}
-
-float applySingleChannel(float value, float inBlack, float inWhite, float gamma, float outBlack, float outWhite) {
-    float inRange = max(inWhite - inBlack, 0.0001);
-    float result = clamp((value - inBlack) / inRange, 0.0, 1.0);
-    result = pow(result, 1.0 / gamma);
-    result = mix(outBlack, outWhite, result);
-    return result;
-}
-
-void main() {
-    vec4 texColor = texture(u_image0, v_texCoord);
-    vec3 color = texColor.rgb;
-    
-    float inBlack = u_float0 / 255.0;
-    float inWhite = u_float1 / 255.0;
-    float gamma = u_float2;
-    float outBlack = u_float3 / 255.0;
-    float outWhite = u_float4 / 255.0;
-    
-    vec3 result;
-    
-    if (u_int0 == 0) {
-        result = applyLevels(color, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 1) {
-        result = color;
-        result.r = applySingleChannel(color.r, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 2) {
-        result = color;
-        result.g = applySingleChannel(color.g, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 3) {
-        result = color;
-        result.b = applySingleChannel(color.b, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else {
-        result = color;
-    }
-    
-    fragColor = vec4(result, texColor.a);
-}

blueprints/.glsl/README.md

@@ -1,28 +0,0 @@
-# GLSL Shader Sources
-
-This folder contains the GLSL fragment shaders extracted from blueprint JSON files for easier editing and version control.
-
-## File Naming Convention
-
-`{Blueprint_Name}_{node_id}.frag`
-
-- **Blueprint_Name**: The JSON filename with spaces/special chars replaced by underscores
-- **node_id**: The GLSLShader node ID within the subgraph
-
-## Usage
-
-```bash
-# Extract shaders from blueprint JSONs to this folder
-python update_blueprints.py extract
-
-# Patch edited shaders back into blueprint JSONs
-python update_blueprints.py patch
-```
-
-## Workflow
-
-1. Run `extract` to pull current shaders from JSONs
-2. Edit `.frag` files
-3. Run `patch` to update the blueprint JSONs
-4. Test
-5. Commit both `.frag` files and updated JSONs

blueprints/.glsl/Sharpen_23.frag

@@ -1,28 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0;  // strength [0.0 – 2.0] typical: 0.3–1.0
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-void main() {
-    vec2 texel = 1.0 / u_resolution;
-    
-    // Sample center and neighbors
-    vec4 center = texture(u_image0, v_texCoord);
-    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));
-    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));
-    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));
-    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));
-    
-    // Edge enhancement (Laplacian)
-    vec4 edges = center * 4.0 - top - bottom - left - right;
-    
-    // Add edges back scaled by strength
-    vec4 sharpened = center + edges * u_float0;
-    
-    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);
-}

blueprints/.glsl/Unsharp_Mask_26.frag

@@ -1,61 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0;  // amount    [0.0 - 3.0]  typical: 0.5-1.5
-uniform float u_float1;  // radius    [0.5 - 10.0] blur radius in pixels
-uniform float u_float2;  // threshold [0.0 - 0.1]  min difference to sharpen
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-float gaussian(float x, float sigma) {
-    return exp(-(x * x) / (2.0 * sigma * sigma));
-}
-
-float getLuminance(vec3 color) {
-    return dot(color, vec3(0.2126, 0.7152, 0.0722));
-}
-
-void main() {
-    vec2 texel = 1.0 / u_resolution;
-    float radius = max(u_float1, 0.5);
-    float amount = u_float0;
-    float threshold = u_float2;
-
-    vec4 original = texture(u_image0, v_texCoord);
-
-    // Gaussian blur for the "unsharp" mask
-    int samples = int(ceil(radius));
-    float sigma = radius / 2.0;
-
-    vec4 blurred = vec4(0.0);
-    float totalWeight = 0.0;
-
-    for (int x = -samples; x <= samples; x++) {
-        for (int y = -samples; y <= samples; y++) {
-            vec2 offset = vec2(float(x), float(y)) * texel;
-            vec4 sample_color = texture(u_image0, v_texCoord + offset);
-
-            float dist = length(vec2(float(x), float(y)));
-            float weight = gaussian(dist, sigma);
-            blurred += sample_color * weight;
-            totalWeight += weight;
-        }
-    }
-    blurred /= totalWeight;
-
-    // Unsharp mask = original - blurred
-    vec3 mask = original.rgb - blurred.rgb;
-
-    // Luminance-based threshold with smooth falloff
-    float lumaDelta = abs(getLuminance(original.rgb) - getLuminance(blurred.rgb));
-    float thresholdScale = smoothstep(0.0, threshold, lumaDelta);
-    mask *= thresholdScale;
-
-    // Sharpen: original + mask * amount
-    vec3 sharpened = original.rgb + mask * amount;
-
-    fragColor0 = vec4(clamp(sharpened, 0.0, 1.0), original.a);
-}

blueprints/.glsl/update_blueprints.py

@@ -1,159 +0,0 @@
-#!/usr/bin/env python3
-"""
-Shader Blueprint Updater
-
-Syncs GLSL shader files between this folder and blueprint JSON files.
-
-File naming convention:
-    {Blueprint Name}_{node_id}.frag
-
-Usage:
-    python update_blueprints.py extract   # Extract shaders from JSONs to here
-    python update_blueprints.py patch     # Patch shaders back into JSONs
-    python update_blueprints.py           # Same as patch (default)
-"""
-
-import json
-import logging
-import sys
-import re
-from pathlib import Path
-
-logging.basicConfig(level=logging.INFO, format='%(message)s')
-logger = logging.getLogger(__name__)
-
-GLSL_DIR = Path(__file__).parent
-BLUEPRINTS_DIR = GLSL_DIR.parent
-
-
-def get_blueprint_files():
-    """Get all blueprint JSON files."""
-    return sorted(BLUEPRINTS_DIR.glob("*.json"))
-
-
-def sanitize_filename(name):
-    """Convert blueprint name to safe filename."""
-    return re.sub(r'[^\w\-]', '_', name)
-
-
-def extract_shaders():
-    """Extract all shaders from blueprint JSONs to this folder."""
-    extracted = 0
-    for json_path in get_blueprint_files():
-        blueprint_name = json_path.stem
-
-        try:
-            with open(json_path, 'r') as f:
-                data = json.load(f)
-        except (json.JSONDecodeError, IOError) as e:
-            logger.warning("Skipping %s: %s", json_path.name, e)
-            continue
-
-        # Find GLSLShader nodes in subgraphs
-        for subgraph in data.get('definitions', {}).get('subgraphs', []):
-            for node in subgraph.get('nodes', []):
-                if node.get('type') == 'GLSLShader':
-                    node_id = node.get('id')
-                    widgets = node.get('widgets_values', [])
-
-                    # Find shader code (first string that looks like GLSL)
-                    for widget in widgets:
-                        if isinstance(widget, str) and widget.startswith('#version'):
-                            safe_name = sanitize_filename(blueprint_name)
-                            frag_name = f"{safe_name}_{node_id}.frag"
-                            frag_path = GLSL_DIR / frag_name
-
-                            with open(frag_path, 'w') as f:
-                                f.write(widget)
-
-                            logger.info("  Extracted: %s", frag_name)
-                            extracted += 1
-                            break
-
-    logger.info("\nExtracted %d shader(s)", extracted)
-
-
-def patch_shaders():
-    """Patch shaders from this folder back into blueprint JSONs."""
-    # Build lookup: blueprint_name -> [(node_id, shader_code), ...]
-    shader_updates = {}
-
-    for frag_path in sorted(GLSL_DIR.glob("*.frag")):
-        # Parse filename: {blueprint_name}_{node_id}.frag
-        parts = frag_path.stem.rsplit('_', 1)
-        if len(parts) != 2:
-            logger.warning("Skipping %s: invalid filename format", frag_path.name)
-            continue
-
-        blueprint_name, node_id_str = parts
-
-        try:
-            node_id = int(node_id_str)
-        except ValueError:
-            logger.warning("Skipping %s: invalid node_id", frag_path.name)
-            continue
-
-        with open(frag_path, 'r') as f:
-            shader_code = f.read()
-
-        if blueprint_name not in shader_updates:
-            shader_updates[blueprint_name] = []
-        shader_updates[blueprint_name].append((node_id, shader_code))
-
-    # Apply updates to JSON files
-    patched = 0
-    for json_path in get_blueprint_files():
-        blueprint_name = sanitize_filename(json_path.stem)
-
-        if blueprint_name not in shader_updates:
-            continue
-
-        try:
-            with open(json_path, 'r') as f:
-                data = json.load(f)
-        except (json.JSONDecodeError, IOError) as e:
-            logger.error("Error reading %s: %s", json_path.name, e)
-            continue
-
-        modified = False
-        for node_id, shader_code in shader_updates[blueprint_name]:
-            # Find the node and update
-            for subgraph in data.get('definitions', {}).get('subgraphs', []):
-                for node in subgraph.get('nodes', []):
-                    if node.get('id') == node_id and node.get('type') == 'GLSLShader':
-                        widgets = node.get('widgets_values', [])
-                        if len(widgets) > 0 and widgets[0] != shader_code:
-                            widgets[0] = shader_code
-                            modified = True
-                            logger.info("  Patched: %s (node %d)", json_path.name, node_id)
-                            patched += 1
-
-        if modified:
-            with open(json_path, 'w') as f:
-                json.dump(data, f)
-
-    if patched == 0:
-        logger.info("No changes to apply.")
-    else:
-        logger.info("\nPatched %d shader(s)", patched)
-
-
-def main():
-    if len(sys.argv) < 2:
-        command = "patch"
-    else:
-        command = sys.argv[1].lower()
-
-    if command == "extract":
-        logger.info("Extracting shaders from blueprints...")
-        extract_shaders()
-    elif command in ("patch", "update", "apply"):
-        logger.info("Patching shaders into blueprints...")
-        patch_shaders()
-    else:
-        logger.info(__doc__)
-        sys.exit(1)
-
-
-if __name__ == "__main__":
-    main()

blueprints/Image Channels.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 29, "last_link_id": 0, "nodes": [{"id": 29, "type": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "pos": [1970, -230], "size": [180, 86], "flags": {}, "order": 5, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": []}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": []}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": []}], "title": "Image Channels", "properties": {"proxyWidgets": []}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 28, "lastLinkId": 39, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Image Channels", "inputNode": {"id": -10, "bounding": [1820, -185, 120, 60]}, "outputNode": {"id": -20, "bounding": [2460, -215, 120, 120]}, "inputs": [{"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe", "name": "images.image0", "type": "IMAGE", "linkIds": [39], "localized_name": "images.image0", "label": "image", "pos": [1920, -165]}], "outputs": [{"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b", "name": "IMAGE0", "type": "IMAGE", "linkIds": [26], "localized_name": "IMAGE0", "label": "R", "pos": [2480, -195]}, {"id": "fb44a77e-0522-43e9-9527-82e7465b3596", "name": "IMAGE1", "type": "IMAGE", "linkIds": [27], "localized_name": "IMAGE1", "label": "G", "pos": [2480, -175]}, {"id": "81460ee6-0131-402a-874f-6bf3001fc4ff", "name": "IMAGE2", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE2", "label": "B", "pos": [2480, -155]}, {"id": "ae690246-80d4-4951-b1d9-9306d8a77417", "name": "IMAGE3", "type": "IMAGE", "linkIds": [29], "localized_name": "IMAGE3", "label": "A", "pos": [2480, -135]}], "widgets": [], "nodes": [{"id": 23, "type": "GLSLShader", "pos": [2000, -330], "size": [400, 172], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 39}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [26]}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": [27]}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": [28]}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": [29]}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n  vec4 color = texture(u_image0, v_texCoord);\n  // Output each channel as grayscale to separate render targets\n  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel\n  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel\n  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel\n  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel\n}\n", "from_input"]}], "groups": [], "links": [{"id": 39, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 26, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}, {"id": 27, "origin_id": 23, "origin_slot": 1, "target_id": -20, "target_slot": 1, "type": "IMAGE"}, {"id": 28, "origin_id": 23, "origin_slot": 2, "target_id": -20, "target_slot": 2, "type": "IMAGE"}, {"id": 29, "origin_id": 23, "origin_slot": 3, "target_id": -20, "target_slot": 3, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}}

blueprints/Prompt Enhance.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 15, "last_link_id": 0, "nodes": [{"id": 15, "type": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "pos": [-1490, 2040], "size": [400, 260], "flags": {}, "order": 0, "mode": 0, "inputs": [{"name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": null}, {"label": "reference images", "name": "images", "type": "IMAGE", "link": null}], "outputs": [{"name": "STRING", "type": "STRING", "links": null}], "title": "Prompt Enhance", "properties": {"proxyWidgets": [["-1", "prompt"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": [""]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 15, "lastLinkId": 14, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Prompt Enhance", "inputNode": {"id": -10, "bounding": [-2170, 2110, 138.876953125, 80]}, "outputNode": {"id": -20, "bounding": [-640, 2110, 120, 60]}, "inputs": [{"id": "aeab7216-00e0-4528-a09b-bba50845c5a6", "name": "prompt", "type": "STRING", "linkIds": [11], "pos": [-2051.123046875, 2130]}, {"id": "7b73fd36-aa31-4771-9066-f6c83879994b", "name": "images", "type": "IMAGE", "linkIds": [14], "label": "reference images", "pos": [-2051.123046875, 2150]}], "outputs": [{"id": "c7b0d930-68a1-48d1-b496-0519e5837064", "name": "STRING", "type": "STRING", "linkIds": [13], "pos": [-620, 2130]}], "widgets": [], "nodes": [{"id": 11, "type": "GeminiNode", "pos": [-1560, 1990], "size": [470, 470], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "shape": 7, "type": "IMAGE", "link": 14}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": null}, {"localized_name": "video", "name": "video", "shape": 7, "type": "VIDEO", "link": null}, {"localized_name": "files", "name": "files", "shape": 7, "type": "GEMINI_INPUT_FILES", "link": null}, {"localized_name": "prompt", "name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": 11}, {"localized_name": "model", "name": "model", "type": "COMBO", "widget": {"name": "model"}, "link": null}, {"localized_name": "seed", "name": "seed", "type": "INT", "widget": {"name": "seed"}, "link": null}, {"localized_name": "system_prompt", "name": "system_prompt", "shape": 7, "type": "STRING", "widget": {"name": "system_prompt"}, "link": null}], "outputs": [{"localized_name": "STRING", "name": "STRING", "type": "STRING", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.14.1", "Node name for S&R": "GeminiNode"}, "widgets_values": ["", "gemini-3-pro-preview", 42, "randomize", "You are an expert in prompt writing.\nBased on the input, rewrite the user's input into a detailed prompt.\nincluding camera settings, lighting, composition, and style.\nReturn the prompt only"], "color": "#432", "bgcolor": "#653"}], "groups": [], "links": [{"id": 11, "origin_id": -10, "origin_slot": 0, "target_id": 11, "target_slot": 4, "type": "STRING"}, {"id": 13, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "STRING"}, {"id": 14, "origin_id": -10, "origin_slot": 1, "target_id": 11, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Text generation/Prompt enhance"}]}, "extra": {}}

blueprints/Sharpen.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 25, "last_link_id": 0, "nodes": [{"id": 25, "type": "621ba4e2-22a8-482d-a369-023753198b7b", "pos": [4610, -790], "size": [230, 58], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "title": "Sharpen", "properties": {"proxyWidgets": [["24", "value"]]}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "621ba4e2-22a8-482d-a369-023753198b7b", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 24, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Sharpen", "inputNode": {"id": -10, "bounding": [4090, -825, 120, 60]}, "outputNode": {"id": -20, "bounding": [5150, -825, 120, 60]}, "inputs": [{"id": "37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7", "name": "images.image0", "type": "IMAGE", "linkIds": [34], "localized_name": "images.image0", "label": "image", "pos": [4190, -805]}], "outputs": [{"id": "e9182b3f-635c-4cd4-a152-4b4be17ae4b9", "name": "IMAGE0", "type": "IMAGE", "linkIds": [35], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [5170, -805]}], "widgets": [], "nodes": [{"id": 24, "type": "PrimitiveFloat", "pos": [4280, -1240], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "strength", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [36]}], "properties": {"Node name for S&R": "PrimitiveFloat", "min": 0, "max": 3, "precision": 2, "step": 0.05}, "widgets_values": [0.5]}, {"id": 23, "type": "GLSLShader", "pos": [4570, -1240], "size": [370, 192], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 34}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 36}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [35]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0;  // strength [0.0 – 2.0] typical: 0.3–1.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n    vec2 texel = 1.0 / u_resolution;\n    \n    // Sample center and neighbors\n    vec4 center = texture(u_image0, v_texCoord);\n    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));\n    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));\n    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));\n    \n    // Edge enhancement (Laplacian)\n    vec4 edges = center * 4.0 - top - bottom - left - right;\n    \n    // Add edges back scaled by strength\n    vec4 sharpened = center + edges * u_float0;\n    \n    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}", "from_input"]}], "groups": [], "links": [{"id": 36, "origin_id": 24, "origin_slot": 0, "target_id": 23, "target_slot": 2, "type": "FLOAT"}, {"id": 34, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 35, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Sharpen"}]}}

blueprints/Video Upscale(GAN x4).json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 13, "last_link_id": 0, "nodes": [{"id": 13, "type": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "pos": [1120, 330], "size": [240, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": null}, {"name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": null}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": []}], "title": "Video Upscale(GAN x4)", "properties": {"proxyWidgets": [["-1", "model_name"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 13, "lastLinkId": 19, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Video Upscale(GAN x4)", "inputNode": {"id": -10, "bounding": [550, 460, 120, 80]}, "outputNode": {"id": -20, "bounding": [1490, 460, 120, 60]}, "inputs": [{"id": "666d633e-93e7-42dc-8d11-2b7b99b0f2a6", "name": "video", "type": "VIDEO", "linkIds": [10], "localized_name": "video", "pos": [650, 480]}, {"id": "2e23a087-caa8-4d65-99e6-662761aa905a", "name": "model_name", "type": "COMBO", "linkIds": [19], "pos": [650, 500]}], "outputs": [{"id": "0c1768ea-3ec2-412f-9af6-8e0fa36dae70", "name": "VIDEO", "type": "VIDEO", "linkIds": [15], "localized_name": "VIDEO", "pos": [1510, 480]}], "widgets": [], "nodes": [{"id": 2, "type": "ImageUpscaleWithModel", "pos": [1110, 450], "size": [320, 46], "flags": {}, "order": 1, "mode": 0, "inputs": [{"localized_name": "upscale_model", "name": "upscale_model", "type": "UPSCALE_MODEL", "link": 1}, {"localized_name": "image", "name": "image", "type": "IMAGE", "link": 14}], "outputs": [{"localized_name": "IMAGE", "name": "IMAGE", "type": "IMAGE", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "ImageUpscaleWithModel"}}, {"id": 11, "type": "CreateVideo", "pos": [1110, 550], "size": [320, 78], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "link": 13}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": 16}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "widget": {"name": "fps"}, "link": 12}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": [15]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "CreateVideo"}, "widgets_values": [30]}, {"id": 10, "type": "GetVideoComponents", "pos": [1110, 330], "size": [320, 70], "flags": {}, "order": 2, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": 10}], "outputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "links": [14]}, {"localized_name": "audio", "name": "audio", "type": "AUDIO", "links": [16]}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "links": [12]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "GetVideoComponents"}}, {"id": 1, "type": "UpscaleModelLoader", "pos": [750, 450], "size": [280, 60], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "model_name", "name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": 19}], "outputs": [{"localized_name": "UPSCALE_MODEL", "name": "UPSCALE_MODEL", "type": "UPSCALE_MODEL", "links": [1]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "UpscaleModelLoader", "models": [{"name": "RealESRGAN_x4plus.safetensors", "url": "https://huggingface.co/Comfy-Org/Real-ESRGAN_repackaged/resolve/main/RealESRGAN_x4plus.safetensors", "directory": "upscale_models"}]}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "groups": [], "links": [{"id": 1, "origin_id": 1, "origin_slot": 0, "target_id": 2, "target_slot": 0, "type": "UPSCALE_MODEL"}, {"id": 14, "origin_id": 10, "origin_slot": 0, "target_id": 2, "target_slot": 1, "type": "IMAGE"}, {"id": 13, "origin_id": 2, "origin_slot": 0, "target_id": 11, "target_slot": 0, "type": "IMAGE"}, {"id": 16, "origin_id": 10, "origin_slot": 1, "target_id": 11, "target_slot": 1, "type": "AUDIO"}, {"id": 12, "origin_id": 10, "origin_slot": 2, "target_id": 11, "target_slot": 2, "type": "FLOAT"}, {"id": 10, "origin_id": -10, "origin_slot": 0, "target_id": 10, "target_slot": 0, "type": "VIDEO"}, {"id": 15, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "VIDEO"}, {"id": 19, "origin_id": -10, "origin_slot": 1, "target_id": 1, "target_slot": 0, "type": "COMBO"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Video generation and editing/Enhance video"}]}, "extra": {}}

comfy/cli_args.py

@@ -49,7 +49,7 @@ parser.add_argument("--temp-directory", type=str, default=None, help="Set the Co
 parser.add_argument("--input-directory", type=str, default=None, help="Set the ComfyUI input directory. Overrides --base-directory.")
 parser.add_argument("--auto-launch", action="store_true", help="Automatically launch ComfyUI in the default browser.")
 parser.add_argument("--disable-auto-launch", action="store_true", help="Disable auto launching the browser.")
-parser.add_argument("--cuda-device", type=int, default=None, metavar="DEVICE_ID", help="Set the id of the cuda device this instance will use. All other devices will not be visible.")
+parser.add_argument("--cuda-device", type=str, default=None, metavar="DEVICE_ID", help="Set the ids of cuda devices this instance will use. All other devices will not be visible.")
 parser.add_argument("--default-device", type=int, default=None, metavar="DEFAULT_DEVICE_ID", help="Set the id of the default device, all other devices will stay visible.")
 cm_group = parser.add_mutually_exclusive_group()
 cm_group.add_argument("--cuda-malloc", action="store_true", help="Enable cudaMallocAsync (enabled by default for torch 2.0 and up).")

comfy/comfy_types/node_typing.py

@@ -176,8 +176,6 @@ class InputTypeOptions(TypedDict):
     """COMBO type only. Specifies the configuration for a multi-select widget.
     Available after ComfyUI frontend v1.13.4
     https://github.com/Comfy-Org/ComfyUI_frontend/pull/2987"""
-    gradient_stops: NotRequired[list[list[float]]]
-    """Gradient color stops for gradientslider display mode. Each stop is [offset, r, g, b] (``FLOAT``)."""
 
 
 class HiddenInputTypeDict(TypedDict):

comfy/controlnet.py

@@ -15,13 +15,14 @@
     You should have received a copy of the GNU General Public License
     along with this program.  If not, see <https://www.gnu.org/licenses/>.
 """
-
+from __future__ import annotations
 
 import torch
 from enum import Enum
 import math
 import os
 import logging
+import copy
 import comfy.utils
 import comfy.model_management
 import comfy.model_detection
@@ -38,7 +39,7 @@ import comfy.ldm.hydit.controlnet
 import comfy.ldm.flux.controlnet
 import comfy.ldm.qwen_image.controlnet
 import comfy.cldm.dit_embedder
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Union
 if TYPE_CHECKING:
     from comfy.hooks import HookGroup
 
@@ -64,6 +65,18 @@ class StrengthType(Enum):
     CONSTANT = 1
     LINEAR_UP = 2
 
+class ControlIsolation:
+    '''Temporarily set a ControlBase object's previous_controlnet to None to prevent cascading calls.'''
+    def __init__(self, control: ControlBase):
+        self.control = control
+        self.orig_previous_controlnet = control.previous_controlnet
+
+    def __enter__(self):
+        self.control.previous_controlnet = None
+
+    def __exit__(self, *args):
+        self.control.previous_controlnet = self.orig_previous_controlnet
+
 class ControlBase:
     def __init__(self):
         self.cond_hint_original = None
@@ -77,7 +90,7 @@ class ControlBase:
         self.compression_ratio = 8
         self.upscale_algorithm = 'nearest-exact'
         self.extra_args = {}
-        self.previous_controlnet = None
+        self.previous_controlnet: Union[ControlBase, None] = None
         self.extra_conds = []
         self.strength_type = StrengthType.CONSTANT
         self.concat_mask = False
@@ -85,6 +98,7 @@ class ControlBase:
         self.extra_concat = None
         self.extra_hooks: HookGroup = None
         self.preprocess_image = lambda a: a
+        self.multigpu_clones: dict[torch.device, ControlBase] = {}
 
     def set_cond_hint(self, cond_hint, strength=1.0, timestep_percent_range=(0.0, 1.0), vae=None, extra_concat=[]):
         self.cond_hint_original = cond_hint
@@ -111,17 +125,38 @@ class ControlBase:
     def cleanup(self):
         if self.previous_controlnet is not None:
             self.previous_controlnet.cleanup()
-
+        for device_cnet in self.multigpu_clones.values():
+            with ControlIsolation(device_cnet):
+                device_cnet.cleanup()
         self.cond_hint = None
         self.extra_concat = None
         self.timestep_range = None
 
     def get_models(self):
         out = []
+        for device_cnet in self.multigpu_clones.values():
+            out += device_cnet.get_models_only_self()
         if self.previous_controlnet is not None:
             out += self.previous_controlnet.get_models()
         return out
 
+    def get_models_only_self(self):
+        'Calls get_models, but temporarily sets previous_controlnet to None.'
+        with ControlIsolation(self):
+            return self.get_models()
+
+    def get_instance_for_device(self, device):
+        'Returns instance of this Control object intended for selected device.'
+        return self.multigpu_clones.get(device, self)
+
+    def deepclone_multigpu(self, load_device, autoregister=False):
+        '''
+        Create deep clone of Control object where model(s) is set to other devices.
+
+        When autoregister is set to True, the deep clone is also added to multigpu_clones dict.
+        '''
+        raise NotImplementedError("Classes inheriting from ControlBase should define their own deepclone_multigpu funtion.")
+
     def get_extra_hooks(self):
         out = []
         if self.extra_hooks is not None:
@@ -130,7 +165,7 @@ class ControlBase:
             out += self.previous_controlnet.get_extra_hooks()
         return out
 
-    def copy_to(self, c):
+    def copy_to(self, c: ControlBase):
         c.cond_hint_original = self.cond_hint_original
         c.strength = self.strength
         c.timestep_percent_range = self.timestep_percent_range
@@ -284,6 +319,14 @@ class ControlNet(ControlBase):
         self.copy_to(c)
         return c
 
+    def deepclone_multigpu(self, load_device, autoregister=False):
+        c = self.copy()
+        c.control_model = copy.deepcopy(c.control_model)
+        c.control_model_wrapped = comfy.model_patcher.ModelPatcher(c.control_model, load_device=load_device, offload_device=comfy.model_management.unet_offload_device())
+        if autoregister:
+            self.multigpu_clones[load_device] = c
+        return c
+
     def get_models(self):
         out = super().get_models()
         out.append(self.control_model_wrapped)
@@ -906,6 +949,14 @@ class T2IAdapter(ControlBase):
         self.copy_to(c)
         return c
 
+    def deepclone_multigpu(self, load_device, autoregister=False):
+        c = self.copy()
+        c.t2i_model = copy.deepcopy(c.t2i_model)
+        c.device = load_device
+        if autoregister:
+            self.multigpu_clones[load_device] = c
+        return c
+
 def load_t2i_adapter(t2i_data, model_options={}): #TODO: model_options
     compression_ratio = 8
     upscale_algorithm = 'nearest-exact'

comfy/ldm/anima/model.py

@@ -179,8 +179,8 @@ class LLMAdapter(nn.Module):
             if source_attention_mask.ndim == 2:
                 source_attention_mask = source_attention_mask.unsqueeze(1).unsqueeze(1)
 
+        x = self.in_proj(self.embed(target_input_ids))
         context = source_hidden_states
-        x = self.in_proj(self.embed(target_input_ids, out_dtype=context.dtype))
         position_ids = torch.arange(x.shape[1], device=x.device).unsqueeze(0)
         position_ids_context = torch.arange(context.shape[1], device=x.device).unsqueeze(0)
         position_embeddings = self.rotary_emb(x, position_ids)

comfy/model_management.py

@@ -15,6 +15,7 @@
     You should have received a copy of the GNU General Public License
     along with this program.  If not, see <https://www.gnu.org/licenses/>.
 """
+from __future__ import annotations
 
 import psutil
 import logging
@@ -35,6 +36,10 @@ import comfy.quant_ops
 import comfy_aimdo.torch
 import comfy_aimdo.model_vbar
 
+from typing import TYPE_CHECKING
+if TYPE_CHECKING:
+    from comfy.model_patcher import ModelPatcher
+
 class VRAMState(Enum):
     DISABLED = 0    #No vram present: no need to move models to vram
     NO_VRAM = 1     #Very low vram: enable all the options to save vram
@@ -200,6 +205,25 @@ def get_torch_device():
         else:
             return torch.device(torch.cuda.current_device())
 
+def get_all_torch_devices(exclude_current=False):
+    global cpu_state
+    devices = []
+    if cpu_state == CPUState.GPU:
+        if is_nvidia():
+            for i in range(torch.cuda.device_count()):
+                devices.append(torch.device(i))
+        elif is_intel_xpu():
+            for i in range(torch.xpu.device_count()):
+                devices.append(torch.device(i))
+        elif is_ascend_npu():
+            for i in range(torch.npu.device_count()):
+                devices.append(torch.device(i))
+    else:
+        devices.append(get_torch_device())
+    if exclude_current:
+        devices.remove(get_torch_device())
+    return devices
+
 def get_total_memory(dev=None, torch_total_too=False):
     global directml_enabled
     if dev is None:
@@ -471,9 +495,13 @@ try:
     logging.info("Device: {}".format(get_torch_device_name(get_torch_device())))
 except:
     logging.warning("Could not pick default device.")
+try:
+    for device in get_all_torch_devices(exclude_current=True):
+        logging.info("Device: {}".format(get_torch_device_name(device)))
+except:
+    pass
 
-
-current_loaded_models = []
+current_loaded_models: list[LoadedModel] = []
 
 def module_size(module):
     module_mem = 0
@@ -484,7 +512,7 @@ def module_size(module):
     return module_mem
 
 class LoadedModel:
-    def __init__(self, model):
+    def __init__(self, model: ModelPatcher):
         self._set_model(model)
         self.device = model.load_device
         self.real_model = None
@@ -492,7 +520,7 @@ class LoadedModel:
         self.model_finalizer = None
         self._patcher_finalizer = None
 
-    def _set_model(self, model):
+    def _set_model(self, model: ModelPatcher):
         self._model = weakref.ref(model)
         if model.parent is not None:
             self._parent_model = weakref.ref(model.parent)
@@ -1713,7 +1741,34 @@ def soft_empty_cache(force=False):
         torch.cuda.ipc_collect()
 
 def unload_all_models():
-    free_memory(1e30, get_torch_device())
+    for device in get_all_torch_devices():
+        free_memory(1e30, device)
+
+def unload_model_and_clones(model: ModelPatcher, unload_additional_models=True, all_devices=False):
+    'Unload only model and its clones - primarily for multigpu cloning purposes.'
+    initial_keep_loaded: list[LoadedModel] = current_loaded_models.copy()
+    additional_models = []
+    if unload_additional_models:
+        additional_models = model.get_nested_additional_models()
+    keep_loaded = []
+    for loaded_model in initial_keep_loaded:
+        if loaded_model.model is not None:
+            if model.clone_base_uuid == loaded_model.model.clone_base_uuid:
+                continue
+            # check additional models if they are a match
+            skip = False
+            for add_model in additional_models:
+                if add_model.clone_base_uuid == loaded_model.model.clone_base_uuid:
+                    skip = True
+                    break
+            if skip:
+                continue
+        keep_loaded.append(loaded_model)
+    if not all_devices:
+        free_memory(1e30, get_torch_device(), keep_loaded)
+    else:
+        for device in get_all_torch_devices():
+            free_memory(1e30, device, keep_loaded)
 
 def debug_memory_summary():
     if is_amd() or is_nvidia():

comfy/model_patcher.py

@@ -23,6 +23,7 @@ import inspect
 import logging
 import math
 import uuid
+import copy
 from typing import Callable, Optional
 
 import torch
@@ -75,12 +76,15 @@ def set_model_options_pre_cfg_function(model_options, pre_cfg_function, disable_
 def create_model_options_clone(orig_model_options: dict):
     return comfy.patcher_extension.copy_nested_dicts(orig_model_options)
 
-def create_hook_patches_clone(orig_hook_patches):
+def create_hook_patches_clone(orig_hook_patches, copy_tuples=False):
     new_hook_patches = {}
     for hook_ref in orig_hook_patches:
         new_hook_patches[hook_ref] = {}
         for k in orig_hook_patches[hook_ref]:
             new_hook_patches[hook_ref][k] = orig_hook_patches[hook_ref][k][:]
+            if copy_tuples:
+                for i in range(len(new_hook_patches[hook_ref][k])):
+                    new_hook_patches[hook_ref][k][i] = tuple(new_hook_patches[hook_ref][k][i])
     return new_hook_patches
 
 def wipe_lowvram_weight(m):
@@ -271,6 +275,10 @@ class ModelPatcher:
         self.is_clip = False
         self.hook_mode = comfy.hooks.EnumHookMode.MaxSpeed
 
+        self.cached_patcher_init: tuple[Callable, tuple] | tuple[Callable, tuple, int] | None = None
+        self.is_multigpu_base_clone = False
+        self.clone_base_uuid = uuid.uuid4()
+
         if not hasattr(self.model, 'model_loaded_weight_memory'):
             self.model.model_loaded_weight_memory = 0
 
@@ -362,18 +370,99 @@ class ModelPatcher:
         n.is_clip = self.is_clip
         n.hook_mode = self.hook_mode
 
+        n.cached_patcher_init = self.cached_patcher_init
+        n.is_multigpu_base_clone = self.is_multigpu_base_clone
+        n.clone_base_uuid = self.clone_base_uuid
+
         for callback in self.get_all_callbacks(CallbacksMP.ON_CLONE):
             callback(self, n)
         return n
 
+    def deepclone_multigpu(self, new_load_device=None, models_cache: dict[uuid.UUID,ModelPatcher]=None):
+        logging.info(f"Creating deepclone of {self.model.__class__.__name__} for {new_load_device if new_load_device else self.load_device}.")
+        comfy.model_management.unload_model_and_clones(self)
+        n = self.clone()
+        # set load device, if present
+        if new_load_device is not None:
+            n.load_device = new_load_device
+        if self.cached_patcher_init is not None:
+            temp_model_patcher: ModelPatcher | list[ModelPatcher] = self.cached_patcher_init[0](*self.cached_patcher_init[1])
+            if len(self.cached_patcher_init) > 2:
+                temp_model_patcher = temp_model_patcher[self.cached_patcher_init[2]]
+            n.model = temp_model_patcher.model
+        else:
+            n.model = copy.deepcopy(n.model)
+        # unlike for normal clone, backup dicts that shared same ref should not;
+        # otherwise, patchers that have deep copies of base models will erroneously influence each other.
+        n.backup = copy.deepcopy(n.backup)
+        n.object_patches_backup = copy.deepcopy(n.object_patches_backup)
+        n.hook_backup = copy.deepcopy(n.hook_backup)
+        # multigpu clone should not have multigpu additional_models entry
+        n.remove_additional_models("multigpu")
+        # multigpu_clone all stored additional_models; make sure circular references are properly handled
+        if models_cache is None:
+            models_cache = {}
+        for key, model_list in n.additional_models.items():
+            for i in range(len(model_list)):
+                add_model = n.additional_models[key][i]
+                if add_model.clone_base_uuid not in models_cache:
+                    models_cache[add_model.clone_base_uuid] = add_model.deepclone_multigpu(new_load_device=new_load_device, models_cache=models_cache)
+                n.additional_models[key][i] = models_cache[add_model.clone_base_uuid]
+        for callback in self.get_all_callbacks(CallbacksMP.ON_DEEPCLONE_MULTIGPU):
+            callback(self, n)
+        return n
+
+    def match_multigpu_clones(self):
+        multigpu_models = self.get_additional_models_with_key("multigpu")
+        if len(multigpu_models) > 0:
+            new_multigpu_models = []
+            for mm in multigpu_models:
+                # clone main model, but bring over relevant props from existing multigpu clone
+                n = self.clone()
+                n.load_device = mm.load_device
+                n.backup = mm.backup
+                n.object_patches_backup = mm.object_patches_backup
+                n.hook_backup = mm.hook_backup
+                n.model = mm.model
+                n.is_multigpu_base_clone = mm.is_multigpu_base_clone
+                n.remove_additional_models("multigpu")
+                orig_additional_models: dict[str, list[ModelPatcher]] = comfy.patcher_extension.copy_nested_dicts(n.additional_models)
+                n.additional_models = comfy.patcher_extension.copy_nested_dicts(mm.additional_models)
+                # figure out which additional models are not present in multigpu clone
+                models_cache = {}
+                for mm_add_model in mm.get_additional_models():
+                    models_cache[mm_add_model.clone_base_uuid] = mm_add_model
+                remove_models_uuids = set(list(models_cache.keys()))
+                for key, model_list in orig_additional_models.items():
+                    for orig_add_model in model_list:
+                        if orig_add_model.clone_base_uuid not in models_cache:
+                            models_cache[orig_add_model.clone_base_uuid] = orig_add_model.deepclone_multigpu(new_load_device=n.load_device, models_cache=models_cache)
+                            existing_list = n.get_additional_models_with_key(key)
+                            existing_list.append(models_cache[orig_add_model.clone_base_uuid])
+                            n.set_additional_models(key, existing_list)
+                        if orig_add_model.clone_base_uuid in remove_models_uuids:
+                            remove_models_uuids.remove(orig_add_model.clone_base_uuid)
+                # remove duplicate additional models
+                for key, model_list in n.additional_models.items():
+                    new_model_list = [x for x in model_list if x.clone_base_uuid not in remove_models_uuids]
+                    n.set_additional_models(key, new_model_list)
+                for callback in self.get_all_callbacks(CallbacksMP.ON_MATCH_MULTIGPU_CLONES):
+                    callback(self, n)
+                new_multigpu_models.append(n)
+            self.set_additional_models("multigpu", new_multigpu_models)
+
     def is_clone(self, other):
         if hasattr(other, 'model') and self.model is other.model:
             return True
         return False
 
-    def clone_has_same_weights(self, clone: 'ModelPatcher'):
-        if not self.is_clone(clone):
-            return False
+    def clone_has_same_weights(self, clone: ModelPatcher, allow_multigpu=False):
+        if allow_multigpu:
+            if self.clone_base_uuid != clone.clone_base_uuid:
+                return False
+        else:
+            if not self.is_clone(clone):
+                return False
 
         if self.current_hooks != clone.current_hooks:
             return False
@@ -1113,7 +1202,7 @@ class ModelPatcher:
         return self.additional_models.get(key, [])
 
     def get_additional_models(self):
-        all_models = []
+        all_models: list[ModelPatcher] = []
         for models in self.additional_models.values():
             all_models.extend(models)
         return all_models
@@ -1167,9 +1256,13 @@ class ModelPatcher:
         for callback in self.get_all_callbacks(CallbacksMP.ON_PRE_RUN):
             callback(self)
 
-    def prepare_state(self, timestep):
+    def prepare_state(self, timestep, model_options, ignore_multigpu=False):
         for callback in self.get_all_callbacks(CallbacksMP.ON_PREPARE_STATE):
-            callback(self, timestep)
+            callback(self, timestep, model_options, ignore_multigpu)
+        if not ignore_multigpu and "multigpu_clones" in model_options:
+            for p in model_options["multigpu_clones"].values():
+                p: ModelPatcher
+                p.prepare_state(timestep, model_options, ignore_multigpu=True)
 
     def restore_hook_patches(self):
         if self.hook_patches_backup is not None:
@@ -1182,12 +1275,18 @@ class ModelPatcher:
     def prepare_hook_patches_current_keyframe(self, t: torch.Tensor, hook_group: comfy.hooks.HookGroup, model_options: dict[str]):
         curr_t = t[0]
         reset_current_hooks = False
+        multigpu_kf_changed_cache = None
         transformer_options = model_options.get("transformer_options", {})
         for hook in hook_group.hooks:
             changed = hook.hook_keyframe.prepare_current_keyframe(curr_t=curr_t, transformer_options=transformer_options)
             # if keyframe changed, remove any cached HookGroups that contain hook with the same hook_ref;
             # this will cause the weights to be recalculated when sampling
             if changed:
+                # cache changed for multigpu usage
+                if "multigpu_clones" in model_options:
+                    if multigpu_kf_changed_cache is None:
+                        multigpu_kf_changed_cache = []
+                    multigpu_kf_changed_cache.append(hook)
                 # reset current_hooks if contains hook that changed
                 if self.current_hooks is not None:
                     for current_hook in self.current_hooks.hooks:
@@ -1199,6 +1298,28 @@ class ModelPatcher:
                         self.cached_hook_patches.pop(cached_group)
         if reset_current_hooks:
             self.patch_hooks(None)
+        if "multigpu_clones" in model_options:
+            for p in model_options["multigpu_clones"].values():
+                p: ModelPatcher
+                p._handle_changed_hook_keyframes(multigpu_kf_changed_cache)
+
+    def _handle_changed_hook_keyframes(self, kf_changed_cache: list[comfy.hooks.Hook]):
+        'Used to handle multigpu behavior inside prepare_hook_patches_current_keyframe.'
+        if kf_changed_cache is None:
+            return
+        reset_current_hooks = False
+        # reset current_hooks if contains hook that changed
+        for hook in kf_changed_cache:
+            if self.current_hooks is not None:
+                for current_hook in self.current_hooks.hooks:
+                    if current_hook == hook:
+                        reset_current_hooks = True
+                        break
+            for cached_group in list(self.cached_hook_patches.keys()):
+                if cached_group.contains(hook):
+                    self.cached_hook_patches.pop(cached_group)
+        if reset_current_hooks:
+            self.patch_hooks(None)
 
     def register_all_hook_patches(self, hooks: comfy.hooks.HookGroup, target_dict: dict[str], model_options: dict=None,
                                   registered: comfy.hooks.HookGroup = None):

comfy/multigpu.py

@@ -0,0 +1,167 @@
+from __future__ import annotations
+import torch
+import logging
+
+from collections import namedtuple
+from typing import TYPE_CHECKING
+if TYPE_CHECKING:
+    from comfy.model_patcher import ModelPatcher
+import comfy.utils
+import comfy.patcher_extension
+import comfy.model_management
+
+
+class GPUOptions:
+    def __init__(self, device_index: int, relative_speed: float):
+        self.device_index = device_index
+        self.relative_speed = relative_speed
+
+    def clone(self):
+        return GPUOptions(self.device_index, self.relative_speed)
+
+    def create_dict(self):
+        return {
+            "relative_speed": self.relative_speed
+        }
+
+class GPUOptionsGroup:
+    def __init__(self):
+        self.options: dict[int, GPUOptions] = {}
+
+    def add(self, info: GPUOptions):
+        self.options[info.device_index] = info
+
+    def clone(self):
+        c = GPUOptionsGroup()
+        for opt in self.options.values():
+            c.add(opt)
+        return c
+
+    def register(self, model: ModelPatcher):
+        opts_dict = {}
+        # get devices that are valid for this model
+        devices: list[torch.device] = [model.load_device]
+        for extra_model in model.get_additional_models_with_key("multigpu"):
+            extra_model: ModelPatcher
+            devices.append(extra_model.load_device)
+        # create dictionary with actual device mapped to its GPUOptions
+        device_opts_list: list[GPUOptions] = []
+        for device in devices:
+            device_opts = self.options.get(device.index, GPUOptions(device_index=device.index, relative_speed=1.0))
+            opts_dict[device] = device_opts.create_dict()
+            device_opts_list.append(device_opts)
+        # make relative_speed relative to 1.0
+        min_speed = min([x.relative_speed for x in device_opts_list])
+        for value in opts_dict.values():
+            value['relative_speed'] /= min_speed
+        model.model_options['multigpu_options'] = opts_dict
+
+
+def create_multigpu_deepclones(model: ModelPatcher, max_gpus: int, gpu_options: GPUOptionsGroup=None, reuse_loaded=False):
+    'Prepare ModelPatcher to contain deepclones of its BaseModel and related properties.'
+    model = model.clone()
+    # check if multigpu is already prepared - get the load devices from them if possible to exclude
+    skip_devices = set()
+    multigpu_models = model.get_additional_models_with_key("multigpu")
+    if len(multigpu_models) > 0:
+        for mm in multigpu_models:
+            skip_devices.add(mm.load_device)
+    skip_devices = list(skip_devices)
+
+    full_extra_devices = comfy.model_management.get_all_torch_devices(exclude_current=True)
+    limit_extra_devices = full_extra_devices[:max_gpus-1]
+    extra_devices = limit_extra_devices.copy()
+    # exclude skipped devices
+    for skip in skip_devices:
+        if skip in extra_devices:
+            extra_devices.remove(skip)
+    # create new deepclones
+    if len(extra_devices) > 0:
+        for device in extra_devices:
+            device_patcher = None
+            if reuse_loaded:
+                # check if there are any ModelPatchers currently loaded that could be referenced here after a clone
+                loaded_models: list[ModelPatcher] = comfy.model_management.loaded_models()
+                for lm in loaded_models:
+                    if lm.model is not None and lm.clone_base_uuid == model.clone_base_uuid and lm.load_device == device:
+                        device_patcher = lm.clone()
+                        logging.info(f"Reusing loaded deepclone of {device_patcher.model.__class__.__name__} for {device}")
+                        break
+            if device_patcher is None:
+                device_patcher = model.deepclone_multigpu(new_load_device=device)
+                device_patcher.is_multigpu_base_clone = True
+            multigpu_models = model.get_additional_models_with_key("multigpu")
+            multigpu_models.append(device_patcher)
+            model.set_additional_models("multigpu", multigpu_models)
+        model.match_multigpu_clones()
+        if gpu_options is None:
+            gpu_options = GPUOptionsGroup()
+        gpu_options.register(model)
+    else:
+        logging.info("No extra torch devices need initialization, skipping initializing MultiGPU Work Units.")
+    # TODO: only keep model clones that don't go 'past' the intended max_gpu count
+    # multigpu_models = model.get_additional_models_with_key("multigpu")
+    # new_multigpu_models = []
+    # for m in multigpu_models:
+    #     if m.load_device in limit_extra_devices:
+    #         new_multigpu_models.append(m)
+    # model.set_additional_models("multigpu", new_multigpu_models)
+    # persist skip_devices for use in sampling code
+    # if len(skip_devices) > 0 or "multigpu_skip_devices" in model.model_options:
+    #     model.model_options["multigpu_skip_devices"] = skip_devices
+    return model
+
+
+LoadBalance = namedtuple('LoadBalance', ['work_per_device', 'idle_time'])
+def load_balance_devices(model_options: dict[str], total_work: int, return_idle_time=False, work_normalized: int=None):
+    'Optimize work assigned to different devices, accounting for their relative speeds and splittable work.'
+    opts_dict = model_options['multigpu_options']
+    devices = list(model_options['multigpu_clones'].keys())
+    speed_per_device = []
+    work_per_device = []
+    # get sum of each device's relative_speed
+    total_speed = 0.0
+    for opts in opts_dict.values():
+        total_speed += opts['relative_speed']
+    # get relative work for each device;
+    # obtained by w = (W*r)/R
+    for device in devices:
+        relative_speed = opts_dict[device]['relative_speed']
+        relative_work = (total_work*relative_speed) / total_speed
+        speed_per_device.append(relative_speed)
+        work_per_device.append(relative_work)
+    # relative work must be expressed in whole numbers, but likely is a decimal;
+    # perform rounding while maintaining total sum equal to total work (sum of relative works)
+    work_per_device = round_preserved(work_per_device)
+    dict_work_per_device = {}
+    for device, relative_work in zip(devices, work_per_device):
+        dict_work_per_device[device] = relative_work
+    if not return_idle_time:
+        return LoadBalance(dict_work_per_device, None)
+    # divide relative work by relative speed to get estimated completion time of said work by each device;
+    # time here is relative and does not correspond to real-world units
+    completion_time = [w/r for w,r in zip(work_per_device, speed_per_device)]
+    # calculate relative time spent by the devices waiting on each other after their work is completed
+    idle_time = abs(min(completion_time) - max(completion_time))
+    # if need to compare work idle time, need to normalize to a common total work
+    if work_normalized:
+        idle_time *= (work_normalized/total_work)
+
+    return LoadBalance(dict_work_per_device, idle_time)
+
+def round_preserved(values: list[float]):
+    'Round all values in a list, preserving the combined sum of values.'
+    # get floor of values; casting to int does it too
+    floored = [int(x) for x in values]
+    total_floored = sum(floored)
+    # get remainder to distribute
+    remainder = round(sum(values)) - total_floored
+    # pair values with fractional portions
+    fractional = [(i, x-floored[i]) for i, x in enumerate(values)]
+    # sort by fractional part in descending order
+    fractional.sort(key=lambda x: x[1], reverse=True)
+    # distribute the remainder
+    for i in range(remainder):
+        index = fractional[i][0]
+        floored[index] += 1
+    return floored

comfy/ops.py

@@ -79,7 +79,7 @@ def cast_to_input(weight, input, non_blocking=False, copy=True):
     return comfy.model_management.cast_to(weight, input.dtype, input.device, non_blocking=non_blocking, copy=copy)
 
 
-def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant):
+def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype):
     offload_stream = None
     xfer_dest = None
 
@@ -170,10 +170,10 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu
             #FIXME: this is not accurate, we need to be sensitive to the compute dtype
             x = lowvram_fn(x)
             if (isinstance(orig, QuantizedTensor) and
-                (want_requant and len(fns) == 0 or update_weight)):
+                (orig.dtype == dtype and len(fns) == 0 or update_weight)):
                 seed = comfy.utils.string_to_seed(s.seed_key)
                 y = QuantizedTensor.from_float(x, s.layout_type, scale="recalculate", stochastic_rounding=seed)
-                if want_requant and len(fns) == 0:
+                if orig.dtype == dtype and len(fns) == 0:
                     #The layer actually wants our freshly saved QT
                     x = y
             elif update_weight:
@@ -194,7 +194,7 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu
     return weight, bias, (offload_stream, device if signature is not None else None, None)
 
 
-def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, offloadable=False, compute_dtype=None, want_requant=False):
+def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, offloadable=False, compute_dtype=None):
     # NOTE: offloadable=False is a a legacy and if you are a custom node author reading this please pass
     # offloadable=True and call uncast_bias_weight() after your last usage of the weight/bias. This
     # will add async-offload support to your cast and improve performance.
@@ -212,7 +212,7 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of
     non_blocking = comfy.model_management.device_supports_non_blocking(device)
 
     if hasattr(s, "_v"):
-        return cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant)
+        return cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype)
 
     if offloadable and (device != s.weight.device or
                         (s.bias is not None and device != s.bias.device)):
@@ -850,8 +850,8 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
             def _forward(self, input, weight, bias):
                 return torch.nn.functional.linear(input, weight, bias)
 
-            def forward_comfy_cast_weights(self, input, compute_dtype=None, want_requant=False):
-                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype, want_requant=want_requant)
+            def forward_comfy_cast_weights(self, input, compute_dtype=None):
+                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype)
                 x = self._forward(input, weight, bias)
                 uncast_bias_weight(self, weight, bias, offload_stream)
                 return x
@@ -881,7 +881,8 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                             scale = comfy.model_management.cast_to_device(scale, input.device, None)
                         input = QuantizedTensor.from_float(input_reshaped, self.layout_type, scale=scale)
 
-                output = self.forward_comfy_cast_weights(input, compute_dtype, want_requant=isinstance(input, QuantizedTensor))
+
+                output = self.forward_comfy_cast_weights(input, compute_dtype)
 
                 # Reshape output back to 3D if input was 3D
                 if reshaped_3d:

comfy/patcher_extension.py

@@ -3,6 +3,8 @@ from typing import Callable
 
 class CallbacksMP:
     ON_CLONE = "on_clone"
+    ON_DEEPCLONE_MULTIGPU = "on_deepclone_multigpu"
+    ON_MATCH_MULTIGPU_CLONES = "on_match_multigpu_clones"
     ON_LOAD = "on_load_after"
     ON_DETACH = "on_detach_after"
     ON_CLEANUP = "on_cleanup"

comfy/quant_ops.py

@@ -20,7 +20,7 @@ try:
         if cuda_version < (13,):
             ck.registry.disable("cuda")
             logging.warning("WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.")
-
+    ck.registry.disable("cuda") # multigpu will not work rn with comfy-kitchen on cuda backend
     ck.registry.disable("triton")
     for k, v in ck.list_backends().items():
         logging.info(f"Found comfy_kitchen backend {k}: {v}")

comfy/sampler_helpers.py

@@ -1,16 +1,17 @@
 from __future__ import annotations
+import torch
 import uuid
 import math
 import collections
 import comfy.model_management
 import comfy.conds
+import comfy.model_patcher
 import comfy.utils
 import comfy.hooks
 import comfy.patcher_extension
 from typing import TYPE_CHECKING
 if TYPE_CHECKING:
     from comfy.model_patcher import ModelPatcher
-    from comfy.model_base import BaseModel
     from comfy.controlnet import ControlBase
 
 def prepare_mask(noise_mask, shape, device):
@@ -106,6 +107,47 @@ def cleanup_additional_models(models):
         if hasattr(m, 'cleanup'):
             m.cleanup()
 
+def preprocess_multigpu_conds(conds: dict[str, list[dict[str]]], model: ModelPatcher, model_options: dict[str]):
+    '''If multigpu acceleration required, creates deepclones of ControlNets and GLIGEN per device.'''
+    multigpu_models: list[ModelPatcher] = model.get_additional_models_with_key("multigpu")
+    if len(multigpu_models) == 0:
+        return
+    extra_devices = [x.load_device for x in multigpu_models]
+    # handle controlnets
+    controlnets: set[ControlBase] = set()
+    for k in conds:
+        for kk in conds[k]:
+            if 'control' in kk:
+                controlnets.add(kk['control'])
+    if len(controlnets) > 0:
+        # first, unload all controlnet clones
+        for cnet in list(controlnets):
+            cnet_models = cnet.get_models()
+            for cm in cnet_models:
+                comfy.model_management.unload_model_and_clones(cm, unload_additional_models=True)
+
+        # next, make sure each controlnet has a deepclone for all relevant devices
+        for cnet in controlnets:
+            curr_cnet = cnet
+            while curr_cnet is not None:
+                for device in extra_devices:
+                    if device not in curr_cnet.multigpu_clones:
+                        curr_cnet.deepclone_multigpu(device, autoregister=True)
+                curr_cnet = curr_cnet.previous_controlnet
+        # since all device clones are now present, recreate the linked list for cloned cnets per device
+        for cnet in controlnets:
+            curr_cnet = cnet
+            while curr_cnet is not None:
+                prev_cnet = curr_cnet.previous_controlnet
+                for device in extra_devices:
+                    device_cnet = curr_cnet.get_instance_for_device(device)
+                    prev_device_cnet = None
+                    if prev_cnet is not None:
+                        prev_device_cnet = prev_cnet.get_instance_for_device(device)
+                    device_cnet.set_previous_controlnet(prev_device_cnet)
+                curr_cnet = prev_cnet
+    # potentially handle gligen - since not widely used, ignored for now
+
 def estimate_memory(model, noise_shape, conds):
     cond_shapes = collections.defaultdict(list)
     cond_shapes_min = {}
@@ -130,7 +172,8 @@ def prepare_sampling(model: ModelPatcher, noise_shape, conds, model_options=None
     return executor.execute(model, noise_shape, conds, model_options=model_options, force_full_load=force_full_load, force_offload=force_offload)
 
 def _prepare_sampling(model: ModelPatcher, noise_shape, conds, model_options=None, force_full_load=False, force_offload=False):
-    real_model: BaseModel = None
+    model.match_multigpu_clones()
+    preprocess_multigpu_conds(conds, model, model_options)
     models, inference_memory = get_additional_models(conds, model.model_dtype())
     models += get_additional_models_from_model_options(model_options)
     models += model.get_nested_additional_models()  # TODO: does this require inference_memory update?
@@ -142,7 +185,7 @@ def _prepare_sampling(model: ModelPatcher, noise_shape, conds, model_options=Non
         memory_required += inference_memory
         minimum_memory_required += inference_memory
     comfy.model_management.load_models_gpu([model] + models, memory_required=memory_required, minimum_memory_required=minimum_memory_required, force_full_load=force_full_load)
-    real_model = model.model
+    real_model: BaseModel = model.model
 
     return real_model, conds, models
 
@@ -188,3 +231,18 @@ def prepare_model_patcher(model: ModelPatcher, conds, model_options: dict):
         comfy.patcher_extension.merge_nested_dicts(to_load_options.setdefault(wc_name, {}), model_options["transformer_options"][wc_name],
                                                     copy_dict1=False)
     return to_load_options
+
+def prepare_model_patcher_multigpu_clones(model_patcher: ModelPatcher, loaded_models: list[ModelPatcher], model_options: dict):
+    '''
+    In case multigpu acceleration is enabled, prep ModelPatchers for each device.
+    '''
+    multigpu_patchers: list[ModelPatcher] = [x for x in loaded_models if x.is_multigpu_base_clone]
+    if len(multigpu_patchers) > 0:
+        multigpu_dict: dict[torch.device, ModelPatcher] = {}
+        multigpu_dict[model_patcher.load_device] = model_patcher
+        for x in multigpu_patchers:
+            x.hook_patches = comfy.model_patcher.create_hook_patches_clone(model_patcher.hook_patches, copy_tuples=True)
+            x.hook_mode = model_patcher.hook_mode # match main model's hook_mode
+            multigpu_dict[x.load_device] = x
+        model_options["multigpu_clones"] = multigpu_dict
+    return multigpu_patchers

comfy/samplers.py

@@ -1,7 +1,9 @@
 from __future__ import annotations
+
+import comfy.model_management
 from .k_diffusion import sampling as k_diffusion_sampling
 from .extra_samplers import uni_pc
-from typing import TYPE_CHECKING, Callable, NamedTuple
+from typing import TYPE_CHECKING, Callable, NamedTuple, Any
 if TYPE_CHECKING:
     from comfy.model_patcher import ModelPatcher
     from comfy.model_base import BaseModel
@@ -19,6 +21,7 @@ import comfy.context_windows
 import comfy.utils
 import scipy.stats
 import numpy
+import threading
 
 
 def add_area_dims(area, num_dims):
@@ -141,7 +144,7 @@ def can_concat_cond(c1, c2):
 
     return cond_equal_size(c1.conditioning, c2.conditioning)
 
-def cond_cat(c_list):
+def cond_cat(c_list, device=None):
     temp = {}
     for x in c_list:
         for k in x:
@@ -153,6 +156,8 @@ def cond_cat(c_list):
     for k in temp:
         conds = temp[k]
         out[k] = conds[0].concat(conds[1:])
+        if device is not None and hasattr(out[k], 'to'):
+            out[k] = out[k].to(device)
 
     return out
 
@@ -212,7 +217,9 @@ def _calc_cond_batch_outer(model: BaseModel, conds: list[list[dict]], x_in: torc
     )
     return executor.execute(model, conds, x_in, timestep, model_options)
 
-def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep, model_options):
+def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
+    if 'multigpu_clones' in model_options:
+        return _calc_cond_batch_multigpu(model, conds, x_in, timestep, model_options)
     out_conds = []
     out_counts = []
     # separate conds by matching hooks
@@ -244,7 +251,7 @@ def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tens
     if has_default_conds:
         finalize_default_conds(model, hooked_to_run, default_conds, x_in, timestep, model_options)
 
-    model.current_patcher.prepare_state(timestep)
+    model.current_patcher.prepare_state(timestep, model_options)
 
     # run every hooked_to_run separately
     for hooks, to_run in hooked_to_run.items():
@@ -345,6 +352,196 @@ def _calc_cond_batch(model: BaseModel, conds: list[list[dict]], x_in: torch.Tens
 
     return out_conds
 
+def _calc_cond_batch_multigpu(model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
+    out_conds = []
+    out_counts = []
+    # separate conds by matching hooks
+    hooked_to_run: dict[comfy.hooks.HookGroup,list[tuple[tuple,int]]] = {}
+    default_conds = []
+    has_default_conds = False
+
+    output_device = x_in.device
+
+    for i in range(len(conds)):
+        out_conds.append(torch.zeros_like(x_in))
+        out_counts.append(torch.ones_like(x_in) * 1e-37)
+
+        cond = conds[i]
+        default_c = []
+        if cond is not None:
+            for x in cond:
+                if 'default' in x:
+                    default_c.append(x)
+                    has_default_conds = True
+                    continue
+                p = get_area_and_mult(x, x_in, timestep)
+                if p is None:
+                    continue
+                if p.hooks is not None:
+                    model.current_patcher.prepare_hook_patches_current_keyframe(timestep, p.hooks, model_options)
+                hooked_to_run.setdefault(p.hooks, list())
+                hooked_to_run[p.hooks] += [(p, i)]
+        default_conds.append(default_c)
+
+    if has_default_conds:
+        finalize_default_conds(model, hooked_to_run, default_conds, x_in, timestep, model_options)
+
+    model.current_patcher.prepare_state(timestep, model_options)
+
+    devices = [dev_m for dev_m in model_options['multigpu_clones'].keys()]
+    device_batched_hooked_to_run: dict[torch.device, list[tuple[comfy.hooks.HookGroup, tuple]]] = {}
+
+    total_conds = 0
+    for to_run in hooked_to_run.values():
+        total_conds += len(to_run)
+    conds_per_device = max(1, math.ceil(total_conds//len(devices)))
+    index_device = 0
+    current_device = devices[index_device]
+    # run every hooked_to_run separately
+    for hooks, to_run in hooked_to_run.items():
+        while len(to_run) > 0:
+            current_device = devices[index_device % len(devices)]
+            batched_to_run = device_batched_hooked_to_run.setdefault(current_device, [])
+            # keep track of conds currently scheduled onto this device
+            batched_to_run_length = 0
+            for btr in batched_to_run:
+                batched_to_run_length += len(btr[1])
+
+            first = to_run[0]
+            first_shape = first[0][0].shape
+            to_batch_temp = []
+            # make sure not over conds_per_device limit when creating temp batch
+            for x in range(len(to_run)):
+                if can_concat_cond(to_run[x][0], first[0]) and len(to_batch_temp) < (conds_per_device - batched_to_run_length):
+                    to_batch_temp += [x]
+
+            to_batch_temp.reverse()
+            to_batch = to_batch_temp[:1]
+
+            free_memory = comfy.model_management.get_free_memory(current_device)
+            for i in range(1, len(to_batch_temp) + 1):
+                batch_amount = to_batch_temp[:len(to_batch_temp)//i]
+                input_shape = [len(batch_amount) * first_shape[0]] + list(first_shape)[1:]
+                if model.memory_required(input_shape) * 1.5 < free_memory:
+                    to_batch = batch_amount
+                    break
+            conds_to_batch = []
+            for x in to_batch:
+                conds_to_batch.append(to_run.pop(x))
+            batched_to_run_length += len(conds_to_batch)
+
+            batched_to_run.append((hooks, conds_to_batch))
+            if batched_to_run_length >= conds_per_device:
+                index_device += 1
+
+    class thread_result(NamedTuple):
+        output: Any
+        mult: Any
+        area: Any
+        batch_chunks: int
+        cond_or_uncond: Any
+        error: Exception = None
+
+    def _handle_batch(device: torch.device, batch_tuple: tuple[comfy.hooks.HookGroup, tuple], results: list[thread_result]):
+        try:
+            model_current: BaseModel = model_options["multigpu_clones"][device].model
+            # run every hooked_to_run separately
+            with torch.no_grad():
+                for hooks, to_batch in batch_tuple:
+                    input_x = []
+                    mult = []
+                    c = []
+                    cond_or_uncond = []
+                    uuids = []
+                    area = []
+                    control: ControlBase = None
+                    patches = None
+                    for x in to_batch:
+                        o = x
+                        p = o[0]
+                        input_x.append(p.input_x)
+                        mult.append(p.mult)
+                        c.append(p.conditioning)
+                        area.append(p.area)
+                        cond_or_uncond.append(o[1])
+                        uuids.append(p.uuid)
+                        control = p.control
+                        patches = p.patches
+
+                    batch_chunks = len(cond_or_uncond)
+                    input_x = torch.cat(input_x).to(device)
+                    c = cond_cat(c, device=device)
+                    timestep_ = torch.cat([timestep.to(device)] * batch_chunks)
+
+                    transformer_options = model_current.current_patcher.apply_hooks(hooks=hooks)
+                    if 'transformer_options' in model_options:
+                        transformer_options = comfy.patcher_extension.merge_nested_dicts(transformer_options,
+                                                                                        model_options['transformer_options'],
+                                                                                        copy_dict1=False)
+
+                    if patches is not None:
+                        transformer_options["patches"] = comfy.patcher_extension.merge_nested_dicts(
+                            transformer_options.get("patches", {}),
+                            patches
+                        )
+
+                    transformer_options["cond_or_uncond"] = cond_or_uncond[:]
+                    transformer_options["uuids"] = uuids[:]
+                    transformer_options["sigmas"] = timestep.to(device)
+                    transformer_options["sample_sigmas"] = transformer_options["sample_sigmas"].to(device)
+                    transformer_options["multigpu_thread_device"] = device
+
+                    cast_transformer_options(transformer_options, device=device)
+                    c['transformer_options'] = transformer_options
+
+                    if control is not None:
+                        device_control = control.get_instance_for_device(device)
+                        c['control'] = device_control.get_control(input_x, timestep_, c, len(cond_or_uncond), transformer_options)
+
+                    if 'model_function_wrapper' in model_options:
+                        output = model_options['model_function_wrapper'](model_current.apply_model, {"input": input_x, "timestep": timestep_, "c": c, "cond_or_uncond": cond_or_uncond}).to(output_device).chunk(batch_chunks)
+                    else:
+                        output = model_current.apply_model(input_x, timestep_, **c).to(output_device).chunk(batch_chunks)
+                    results.append(thread_result(output, mult, area, batch_chunks, cond_or_uncond))
+        except Exception as e:
+            results.append(thread_result(None, None, None, None, None, error=e))
+            raise
+
+
+    results: list[thread_result] = []
+    threads: list[threading.Thread] = []
+    for device, batch_tuple in device_batched_hooked_to_run.items():
+        new_thread = threading.Thread(target=_handle_batch, args=(device, batch_tuple, results))
+        threads.append(new_thread)
+        new_thread.start()
+
+    for thread in threads:
+        thread.join()
+
+    for output, mult, area, batch_chunks, cond_or_uncond, error in results:
+        if error is not None:
+            raise error
+        for o in range(batch_chunks):
+            cond_index = cond_or_uncond[o]
+            a = area[o]
+            if a is None:
+                out_conds[cond_index] += output[o] * mult[o]
+                out_counts[cond_index] += mult[o]
+            else:
+                out_c = out_conds[cond_index]
+                out_cts = out_counts[cond_index]
+                dims = len(a) // 2
+                for i in range(dims):
+                    out_c = out_c.narrow(i + 2, a[i + dims], a[i])
+                    out_cts = out_cts.narrow(i + 2, a[i + dims], a[i])
+                out_c += output[o] * mult[o]
+                out_cts += mult[o]
+
+    for i in range(len(out_conds)):
+        out_conds[i] /= out_counts[i]
+
+    return out_conds
+
 def calc_cond_uncond_batch(model, cond, uncond, x_in, timestep, model_options): #TODO: remove
     logging.warning("WARNING: The comfy.samplers.calc_cond_uncond_batch function is deprecated please use the calc_cond_batch one instead.")
     return tuple(calc_cond_batch(model, [cond, uncond], x_in, timestep, model_options))
@@ -649,6 +846,8 @@ def pre_run_control(model, conds):
         percent_to_timestep_function = lambda a: s.percent_to_sigma(a)
         if 'control' in x:
             x['control'].pre_run(model, percent_to_timestep_function)
+            for device_cnet in x['control'].multigpu_clones.values():
+                device_cnet.pre_run(model, percent_to_timestep_function)
 
 def apply_empty_x_to_equal_area(conds, uncond, name, uncond_fill_func):
     cond_cnets = []
@@ -891,7 +1090,9 @@ def cast_to_load_options(model_options: dict[str], device=None, dtype=None):
     to_load_options = model_options.get("to_load_options", None)
     if to_load_options is None:
         return
+    cast_transformer_options(to_load_options, device, dtype)
 
+def cast_transformer_options(transformer_options: dict[str], device=None, dtype=None):
     casts = []
     if device is not None:
         casts.append(device)
@@ -900,18 +1101,17 @@ def cast_to_load_options(model_options: dict[str], device=None, dtype=None):
     # if nothing to apply, do nothing
     if len(casts) == 0:
         return
-
     # try to call .to on patches
-    if "patches" in to_load_options:
-        patches = to_load_options["patches"]
+    if "patches" in transformer_options:
+        patches = transformer_options["patches"]
         for name in patches:
             patch_list = patches[name]
             for i in range(len(patch_list)):
                 if hasattr(patch_list[i], "to"):
                     for cast in casts:
                         patch_list[i] = patch_list[i].to(cast)
-    if "patches_replace" in to_load_options:
-        patches = to_load_options["patches_replace"]
+    if "patches_replace" in transformer_options:
+        patches = transformer_options["patches_replace"]
         for name in patches:
             patch_list = patches[name]
             for k in patch_list:
@@ -921,8 +1121,8 @@ def cast_to_load_options(model_options: dict[str], device=None, dtype=None):
     # try to call .to on any wrappers/callbacks
     wrappers_and_callbacks = ["wrappers", "callbacks"]
     for wc_name in wrappers_and_callbacks:
-        if wc_name in to_load_options:
-            wc: dict[str, list] = to_load_options[wc_name]
+        if wc_name in transformer_options:
+            wc: dict[str, list] = transformer_options[wc_name]
             for wc_dict in wc.values():
                 for wc_list in wc_dict.values():
                     for i in range(len(wc_list)):
@@ -930,7 +1130,6 @@ def cast_to_load_options(model_options: dict[str], device=None, dtype=None):
                             for cast in casts:
                                 wc_list[i] = wc_list[i].to(cast)
 
-
 class CFGGuider:
     def __init__(self, model_patcher: ModelPatcher):
         self.model_patcher = model_patcher
@@ -983,6 +1182,8 @@ class CFGGuider:
         self.inner_model, self.conds, self.loaded_models = comfy.sampler_helpers.prepare_sampling(self.model_patcher, noise.shape, self.conds, self.model_options)
         device = self.model_patcher.load_device
 
+        multigpu_patchers = comfy.sampler_helpers.prepare_model_patcher_multigpu_clones(self.model_patcher, self.loaded_models, self.model_options)
+
         noise = noise.to(device)
         latent_image = latent_image.to(device)
         sigmas = sigmas.to(device)
@@ -990,9 +1191,13 @@ class CFGGuider:
 
         try:
             self.model_patcher.pre_run()
+            for multigpu_patcher in multigpu_patchers:
+                multigpu_patcher.pre_run()
             output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
         finally:
             self.model_patcher.cleanup()
+            for multigpu_patcher in multigpu_patchers:
+                multigpu_patcher.cleanup()
 
         comfy.sampler_helpers.cleanup_models(self.conds, self.loaded_models)
         del self.inner_model

comfy/sd.py

@@ -423,17 +423,6 @@ class CLIP:
     def get_key_patches(self):
         return self.patcher.get_key_patches()
 
-    def generate(self, tokens, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.95, min_p=0.0, repetition_penalty=1.0, seed=None):
-        self.cond_stage_model.reset_clip_options()
-
-        self.load_model()
-        self.cond_stage_model.set_clip_options({"layer": None})
-        self.cond_stage_model.set_clip_options({"execution_device": self.patcher.load_device})
-        return self.cond_stage_model.generate(tokens, do_sample=do_sample, max_length=max_length, temperature=temperature, top_k=top_k, top_p=top_p, min_p=min_p, repetition_penalty=repetition_penalty, seed=seed)
-
-    def decode(self, token_ids, skip_special_tokens=True):
-        return self.tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)
-
 class VAE:
     def __init__(self, sd=None, device=None, config=None, dtype=None, metadata=None):
         if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format
@@ -1193,7 +1182,6 @@ class TEModel(Enum):
     JINA_CLIP_2 = 19
     QWEN3_8B = 20
     QWEN3_06B = 21
-    GEMMA_3_4B_VISION = 22
 
 
 def detect_te_model(sd):
@@ -1222,10 +1210,7 @@ def detect_te_model(sd):
         if 'model.layers.47.self_attn.q_norm.weight' in sd:
             return TEModel.GEMMA_3_12B
         if 'model.layers.0.self_attn.q_norm.weight' in sd:
-            if 'vision_model.embeddings.patch_embedding.weight' in sd:
-                return TEModel.GEMMA_3_4B_VISION
-            else:
-                return TEModel.GEMMA_3_4B
+            return TEModel.GEMMA_3_4B
         return TEModel.GEMMA_2_2B
     if 'model.layers.0.self_attn.k_proj.bias' in sd:
         weight = sd['model.layers.0.self_attn.k_proj.bias']
@@ -1285,8 +1270,6 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
         else:
             if "text_projection" in clip_data[i]:
                 clip_data[i]["text_projection.weight"] = clip_data[i]["text_projection"].transpose(0, 1) #old models saved with the CLIPSave node
-        if "lm_head.weight" in clip_data[i]:
-            clip_data[i]["model.lm_head.weight"] = clip_data[i].pop("lm_head.weight") # prefix missing in some models
 
     tokenizer_data = {}
     clip_target = EmptyClass()
@@ -1352,14 +1335,6 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
             clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data), model_type="gemma3_4b")
             clip_target.tokenizer = comfy.text_encoders.lumina2.NTokenizer
             tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
-        elif te_model == TEModel.GEMMA_3_4B_VISION:
-            clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data), model_type="gemma3_4b_vision")
-            clip_target.tokenizer = comfy.text_encoders.lumina2.NTokenizer
-            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
-        elif te_model == TEModel.GEMMA_3_12B:
-            clip_target.clip = comfy.text_encoders.lt.gemma3_te(**llama_detect(clip_data))
-            clip_target.tokenizer = comfy.text_encoders.lt.Gemma3_12BTokenizer
-            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
         elif te_model == TEModel.LLAMA3_8:
             clip_target.clip = comfy.text_encoders.hidream.hidream_clip(**llama_detect(clip_data),
                                                                         clip_l=False, clip_g=False, t5=False, llama=True, dtype_t5=None)
@@ -1535,6 +1510,7 @@ def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, o
     out = load_state_dict_guess_config(sd, output_vae, output_clip, output_clipvision, embedding_directory, output_model, model_options, te_model_options=te_model_options, metadata=metadata)
     if out is None:
         raise RuntimeError("ERROR: Could not detect model type of: {}\n{}".format(ckpt_path, model_detection_error_hint(ckpt_path, sd)))
+    out[0].cached_patcher_init = (load_checkpoint_guess_config, (ckpt_path, False, False, False, embedding_directory, output_model, model_options, te_model_options), 0)
     return out
 
 def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}, metadata=None):
@@ -1736,6 +1712,7 @@ def load_diffusion_model(unet_path, model_options={}):
     if model is None:
         logging.error("ERROR UNSUPPORTED DIFFUSION MODEL {}".format(unet_path))
         raise RuntimeError("ERROR: Could not detect model type of: {}\n{}".format(unet_path, model_detection_error_hint(unet_path, sd)))
+    model.cached_patcher_init = (load_diffusion_model, (unet_path, model_options))
     return model
 
 def load_unet(unet_path, dtype=None):

comfy/sd1_clip.py

@@ -308,15 +308,6 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
     def load_sd(self, sd):
         return self.transformer.load_state_dict(sd, strict=False, assign=getattr(self, "can_assign_sd", False))
 
-    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
-        if isinstance(tokens, dict):
-            tokens_only = next(iter(tokens.values())) # todo: get this better?
-        else:
-            tokens_only = tokens
-        tokens_only = [[t[0] for t in b] for b in tokens_only]
-        embeds = self.process_tokens(tokens_only, device=self.execution_device)[0]
-        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed)
-
 def parse_parentheses(string):
     result = []
     current_item = ""
@@ -573,8 +564,6 @@ class SDTokenizer:
         min_length = tokenizer_options.get("{}_min_length".format(self.embedding_key), self.min_length)
         min_padding = tokenizer_options.get("{}_min_padding".format(self.embedding_key), self.min_padding)
 
-        min_length = kwargs.get("min_length", min_length)
-
         text = escape_important(text)
         if kwargs.get("disable_weights", self.disable_weights):
             parsed_weights = [(text, 1.0)]
@@ -674,9 +663,6 @@ class SDTokenizer:
     def state_dict(self):
         return {}
 
-    def decode(self, token_ids, skip_special_tokens=True):
-        return self.tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)
-
 class SD1Tokenizer:
     def __init__(self, embedding_directory=None, tokenizer_data={}, clip_name="l", tokenizer=SDTokenizer, name=None):
         if name is not None:
@@ -700,9 +686,6 @@ class SD1Tokenizer:
     def state_dict(self):
         return getattr(self, self.clip).state_dict()
 
-    def decode(self, token_ids, skip_special_tokens=True):
-        return getattr(self, self.clip).decode(token_ids, skip_special_tokens=skip_special_tokens)
-
 class SD1CheckpointClipModel(SDClipModel):
     def __init__(self, device="cpu", dtype=None, model_options={}):
         super().__init__(device=device, return_projected_pooled=False, dtype=dtype, model_options=model_options)
@@ -739,6 +722,3 @@ class SD1ClipModel(torch.nn.Module):
 
     def load_sd(self, sd):
         return getattr(self, self.clip).load_sd(sd)
-
-    def generate(self, tokens, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.95, min_p=0.0, repetition_penalty=1.0, seed=None):
-        return getattr(self, self.clip).generate(tokens, do_sample=do_sample, max_length=max_length, temperature=temperature, top_k=top_k, top_p=top_p, min_p=min_p, repetition_penalty=repetition_penalty, seed=seed)

comfy/text_encoders/anima.py

@@ -33,8 +33,6 @@ class AnimaTokenizer:
     def state_dict(self):
         return {}
 
-    def decode(self, token_ids, **kwargs):
-        return self.qwen3_06b.decode(token_ids, **kwargs)
 
 class Qwen3_06BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="last", layer_idx=None, dtype=None, attention_mask=True, model_options={}):

comfy/text_encoders/llama.py

@@ -3,8 +3,6 @@ import torch.nn as nn
 from dataclasses import dataclass
 from typing import Optional, Any, Tuple
 import math
-from tqdm import tqdm
-import comfy.utils
 
 from comfy.ldm.modules.attention import optimized_attention_for_device
 import comfy.model_management
@@ -105,7 +103,6 @@ class Qwen3_06BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_06B_ACE15_Config:
@@ -129,7 +126,6 @@ class Qwen3_06B_ACE15_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_2B_ACE15_lm_Config:
@@ -153,7 +149,6 @@ class Qwen3_2B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4B_ACE15_lm_Config:
@@ -177,7 +172,6 @@ class Qwen3_4B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4BConfig:
@@ -201,7 +195,6 @@ class Qwen3_4BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_8BConfig:
@@ -225,7 +218,6 @@ class Qwen3_8BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Ovis25_2BConfig:
@@ -296,7 +288,6 @@ class Gemma2_2B_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [1]
 
 @dataclass
 class Gemma3_4B_Config:
@@ -321,14 +312,6 @@ class Gemma3_4B_Config:
     rope_scale = [8.0, 1.0]
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [1, 106]
-
-GEMMA3_VISION_CONFIG = {"num_channels": 3, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "model_type": "siglip_vision_model", "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 14}
-
-@dataclass
-class Gemma3_4B_Vision_Config(Gemma3_4B_Config):
-    vision_config = GEMMA3_VISION_CONFIG
-    mm_tokens_per_image = 256
 
 @dataclass
 class Gemma3_12B_Config:
@@ -353,9 +336,8 @@ class Gemma3_12B_Config:
     rope_scale = [8.0, 1.0]
     final_norm: bool = True
     lm_head: bool = False
-    vision_config = GEMMA3_VISION_CONFIG
+    vision_config = {"num_channels": 3, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "model_type": "siglip_vision_model", "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 14}
     mm_tokens_per_image = 256
-    stop_tokens = [1, 106]
 
 class RMSNorm(nn.Module):
     def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None):
@@ -459,10 +441,8 @@ class Attention(nn.Module):
         freqs_cis: Optional[torch.Tensor] = None,
         optimized_attention=None,
         past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
-        sliding_window: Optional[int] = None,
     ):
         batch_size, seq_length, _ = hidden_states.shape
-
         xq = self.q_proj(hidden_states)
         xk = self.k_proj(hidden_states)
         xv = self.v_proj(hidden_states)
@@ -497,11 +477,6 @@ class Attention(nn.Module):
             else:
                 present_key_value = (xk, xv, index + num_tokens)
 
-            if sliding_window is not None and xk.shape[2] > sliding_window:
-                xk = xk[:, :, -sliding_window:]
-                xv = xv[:, :, -sliding_window:]
-                attention_mask = attention_mask[..., -sliding_window:] if attention_mask is not None else None
-
         xk = xk.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
         xv = xv.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
 
@@ -584,12 +559,10 @@ class TransformerBlockGemma2(nn.Module):
         optimized_attention=None,
         past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
     ):
-        sliding_window = None
         if self.transformer_type == 'gemma3':
             if self.sliding_attention:
-                sliding_window = self.sliding_attention
                 if x.shape[1] > self.sliding_attention:
-                    sliding_mask = torch.full((x.shape[1], x.shape[1]), torch.finfo(x.dtype).min, device=x.device, dtype=x.dtype)
+                    sliding_mask = torch.full((x.shape[1], x.shape[1]), float("-inf"), device=x.device, dtype=x.dtype)
                     sliding_mask.tril_(diagonal=-self.sliding_attention)
                     if attention_mask is not None:
                         attention_mask = attention_mask + sliding_mask
@@ -608,7 +581,6 @@ class TransformerBlockGemma2(nn.Module):
             freqs_cis=freqs_cis,
             optimized_attention=optimized_attention,
             past_key_value=past_key_value,
-            sliding_window=sliding_window,
         )
 
         x = self.post_attention_layernorm(x)
@@ -793,107 +765,6 @@ class BaseLlama:
     def forward(self, input_ids, *args, **kwargs):
         return self.model(input_ids, *args, **kwargs)
 
-class BaseGenerate:
-    def logits(self, x):
-        input = x[:, -1:]
-        if hasattr(self.model, "lm_head"):
-            module = self.model.lm_head
-        else:
-            module = self.model.embed_tokens
-
-        offload_stream = None
-        if module.comfy_cast_weights:
-            weight, _, offload_stream = comfy.ops.cast_bias_weight(module, input, offloadable=True)
-        else:
-            weight = self.model.embed_tokens.weight.to(x)
-
-        x = torch.nn.functional.linear(input, weight, None)
-
-        comfy.ops.uncast_bias_weight(module, weight, None, offload_stream)
-        return x
-
-    def generate(self, embeds=None, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.9, min_p=0.0, repetition_penalty=1.0, seed=42, stop_tokens=None, initial_tokens=[], execution_dtype=None, min_tokens=0):
-        device = embeds.device
-        model_config = self.model.config
-
-        if stop_tokens is None:
-            stop_tokens = self.model.config.stop_tokens
-
-        if execution_dtype is None:
-            if comfy.model_management.should_use_bf16(device):
-                execution_dtype = torch.bfloat16
-            else:
-                execution_dtype = torch.float32
-        embeds = embeds.to(execution_dtype)
-
-        if embeds.ndim == 2:
-            embeds = embeds.unsqueeze(0)
-
-        past_key_values = [] #kv_cache init
-        max_cache_len = embeds.shape[1] + max_length
-        for x in range(model_config.num_hidden_layers):
-            past_key_values.append((torch.empty([embeds.shape[0], model_config.num_key_value_heads, max_cache_len, model_config.head_dim], device=device, dtype=execution_dtype),
-                                    torch.empty([embeds.shape[0], model_config.num_key_value_heads, max_cache_len, model_config.head_dim], device=device, dtype=execution_dtype), 0))
-
-        generator = torch.Generator(device=device).manual_seed(seed) if do_sample else None
-
-        generated_token_ids = []
-        pbar = comfy.utils.ProgressBar(max_length)
-
-        # Generation loop
-        for step in tqdm(range(max_length), desc="Generating tokens"):
-            x, _, past_key_values = self.model.forward(None, embeds=embeds, attention_mask=None, past_key_values=past_key_values)
-            logits = self.logits(x)[:, -1]
-            next_token = self.sample_token(logits, temperature, top_k, top_p, min_p, repetition_penalty, initial_tokens + generated_token_ids, generator, do_sample=do_sample)
-            token_id = next_token[0].item()
-            generated_token_ids.append(token_id)
-
-            embeds = self.model.embed_tokens(next_token).to(execution_dtype)
-            pbar.update(1)
-
-            if token_id in stop_tokens:
-                break
-
-        return generated_token_ids
-
-    def sample_token(self, logits, temperature, top_k, top_p, min_p, repetition_penalty, token_history, generator, do_sample=True):
-
-        if not do_sample or temperature == 0.0:
-            return torch.argmax(logits, dim=-1, keepdim=True)
-
-        # Sampling mode
-        if repetition_penalty != 1.0:
-            for i in range(logits.shape[0]):
-                for token_id in set(token_history):
-                    logits[i, token_id] *= repetition_penalty if logits[i, token_id] < 0 else 1/repetition_penalty
-
-        if temperature != 1.0:
-            logits = logits / temperature
-
-        if top_k > 0:
-            indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
-            logits[indices_to_remove] = torch.finfo(logits.dtype).min
-
-        if min_p > 0.0:
-            probs_before_filter = torch.nn.functional.softmax(logits, dim=-1)
-            top_probs, _ = probs_before_filter.max(dim=-1, keepdim=True)
-            min_threshold = min_p * top_probs
-            indices_to_remove = probs_before_filter < min_threshold
-            logits[indices_to_remove] = torch.finfo(logits.dtype).min
-
-        if top_p < 1.0:
-            sorted_logits, sorted_indices = torch.sort(logits, descending=True)
-            cumulative_probs = torch.cumsum(torch.nn.functional.softmax(sorted_logits, dim=-1), dim=-1)
-            sorted_indices_to_remove = cumulative_probs > top_p
-            sorted_indices_to_remove[..., 0] = False
-            indices_to_remove = torch.zeros_like(logits, dtype=torch.bool)
-            indices_to_remove.scatter_(1, sorted_indices, sorted_indices_to_remove)
-            logits[indices_to_remove] = torch.finfo(logits.dtype).min
-
-        probs = torch.nn.functional.softmax(logits, dim=-1)
-
-        return torch.multinomial(probs, num_samples=1, generator=generator)
-
 class BaseQwen3:
     def logits(self, x):
         input = x[:, -1:]
@@ -937,7 +808,7 @@ class Qwen25_3B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_06B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_06B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_06BConfig(**config_dict)
@@ -964,7 +835,7 @@ class Qwen3_2B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_4B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_4B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_4BConfig(**config_dict)
@@ -982,7 +853,7 @@ class Qwen3_4B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_8B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_8B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_8BConfig(**config_dict)
@@ -1000,7 +871,7 @@ class Ovis25_2B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module):
+class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen25_7BVLI_Config(**config_dict)
@@ -1010,9 +881,6 @@ class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module):
         self.visual = qwen_vl.Qwen2VLVisionTransformer(hidden_size=1280, output_hidden_size=config.hidden_size, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-        # todo: should this be tied or not?
-        #self.lm_head = operations.Linear(config.hidden_size, config.vocab_size, bias=False, device=device, dtype=dtype)
-
     def preprocess_embed(self, embed, device):
         if embed["type"] == "image":
             image, grid = qwen_vl.process_qwen2vl_images(embed["data"])
@@ -1046,7 +914,7 @@ class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module):
 
         return super().forward(x, attention_mask=attention_mask, embeds=embeds, num_tokens=num_tokens, intermediate_output=intermediate_output, final_layer_norm_intermediate=final_layer_norm_intermediate, dtype=dtype, position_ids=position_ids)
 
-class Gemma2_2B(BaseLlama, BaseGenerate, torch.nn.Module):
+class Gemma2_2B(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma2_2B_Config(**config_dict)
@@ -1055,7 +923,7 @@ class Gemma2_2B(BaseLlama, BaseGenerate, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Gemma3_4B(BaseLlama, BaseGenerate, torch.nn.Module):
+class Gemma3_4B(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma3_4B_Config(**config_dict)
@@ -1064,25 +932,7 @@ class Gemma3_4B(BaseLlama, BaseGenerate, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Gemma3_4B_Vision(BaseLlama, BaseGenerate, torch.nn.Module):
-    def __init__(self, config_dict, dtype, device, operations):
-        super().__init__()
-        config = Gemma3_4B_Vision_Config(**config_dict)
-        self.num_layers = config.num_hidden_layers
-
-        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
-        self.dtype = dtype
-        self.multi_modal_projector = Gemma3MultiModalProjector(config, dtype, device, operations)
-        self.vision_model = comfy.clip_model.CLIPVision(config.vision_config, dtype, device, operations)
-        self.image_size = config.vision_config["image_size"]
-
-    def preprocess_embed(self, embed, device):
-        if embed["type"] == "image":
-            image = comfy.clip_model.clip_preprocess(embed["data"], size=self.image_size, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5], crop=True)
-            return self.multi_modal_projector(self.vision_model(image.to(device, dtype=torch.float32))[0]), None
-        return None, None
-
-class Gemma3_12B(BaseLlama, BaseGenerate, torch.nn.Module):
+class Gemma3_12B(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma3_12B_Config(**config_dict)

comfy/text_encoders/lt.py

@@ -6,7 +6,6 @@ import comfy.text_encoders.genmo
 from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector
 import torch
 import comfy.utils
-import math
 
 class T5XXLTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
@@ -23,79 +22,40 @@ def ltxv_te(*args, **kwargs):
     return comfy.text_encoders.genmo.mochi_te(*args, **kwargs)
 
 
-class Gemma3_Tokenizer():
-    def state_dict(self):
-        return {"spiece_model": self.tokenizer.serialize_model()}
-
-    def tokenize_with_weights(self, text, return_word_ids=False, image=None, llama_template=None, skip_template=True, **kwargs):
-        self.llama_template = "<start_of_turn>system\nYou are a helpful assistant.<end_of_turn>\n<start_of_turn>user\n{}<end_of_turn>\n<start_of_turn>model\n"
-        self.llama_template_images = "<start_of_turn>system\nYou are a helpful assistant.<end_of_turn>\n<start_of_turn>user\n\n<image_soft_token>{}<end_of_turn>\n\n<start_of_turn>model\n"
-
-        if image is None:
-            images = []
-        else:
-            samples = image.movedim(-1, 1)
-            total = int(896 * 896)
-
-            scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
-            width = round(samples.shape[3] * scale_by)
-            height = round(samples.shape[2] * scale_by)
-
-            s = comfy.utils.common_upscale(samples, width, height, "area", "disabled").movedim(1, -1)
-            images = [s[:, :, :, :3]]
-
-        if text.startswith('<start_of_turn>'):
-            skip_template = True
-
-        if skip_template:
-            llama_text = text
-        else:
-            if llama_template is None:
-                if len(images) > 0:
-                    llama_text = self.llama_template_images.format(text)
-                else:
-                    llama_text = self.llama_template.format(text)
-            else:
-                llama_text = llama_template.format(text)
-
-        text_tokens = super().tokenize_with_weights(llama_text, return_word_ids)
-
-        if len(images) > 0:
-            embed_count = 0
-            for r in text_tokens:
-                for i, token in enumerate(r):
-                    if token[0] == 262144 and embed_count < len(images):
-                        r[i] = ({"type": "image", "data": images[embed_count]},) + token[1:]
-                        embed_count += 1
-        return text_tokens
-
-class Gemma3_12BTokenizer(Gemma3_Tokenizer, sd1_clip.SDTokenizer):
+class Gemma3_12BTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        special_tokens = {"<image_soft_token>": 262144, "<end_of_turn>": 106}
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=3840, embedding_key='gemma3_12b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=512, pad_left=True, disable_weights=True, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, tokenizer_data=tokenizer_data)
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=3840, embedding_key='gemma3_12b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=512, pad_left=True, disable_weights=True, tokenizer_args={"add_bos": True, "add_eos": False}, tokenizer_data=tokenizer_data)
 
+    def state_dict(self):
+        return {"spiece_model": self.tokenizer.serialize_model()}
 
 class LTXAVGemmaTokenizer(sd1_clip.SD1Tokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="gemma3_12b", tokenizer=Gemma3_12BTokenizer)
 
-
 class Gemma3_12BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="all", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
         llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
         if llama_quantization_metadata is not None:
             model_options = model_options.copy()
             model_options["quantization_metadata"] = llama_quantization_metadata
-        self.dtypes = set()
-        self.dtypes.add(dtype)
+
         super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma3_12B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
-    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
-        tokens_only = [[t[0] for t in b] for b in tokens]
-        embeds, _, _, embeds_info = self.process_tokens(tokens_only, self.execution_device)
-        comfy.utils.normalize_image_embeddings(embeds, embeds_info, self.transformer.model.config.hidden_size ** 0.5)
-        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[106])  # 106 is <end_of_turn>
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template="{}", image_embeds=None, **kwargs):
+        text = llama_template.format(text)
+        text_tokens = super().tokenize_with_weights(text, return_word_ids)
+        embed_count = 0
+        for k in text_tokens:
+            tt = text_tokens[k]
+            for r in tt:
+                for i in range(len(r)):
+                    if r[i][0] == 262144:
+                        if image_embeds is not None and embed_count < image_embeds.shape[0]:
+                            r[i] = ({"type": "embedding", "data": image_embeds[embed_count], "original_type": "image"},) + r[i][1:]
+                            embed_count += 1
+        return text_tokens
 
 class LTXAVTEModel(torch.nn.Module):
     def __init__(self, dtype_llama=None, device="cpu", dtype=None, model_options={}):
@@ -152,9 +112,6 @@ class LTXAVTEModel(torch.nn.Module):
 
         return out.to(out_device), pooled
 
-    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
-        return self.gemma3_12b.generate(tokens["gemma3_12b"], do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed)
-
     def load_sd(self, sd):
         if "model.layers.47.self_attn.q_norm.weight" in sd:
             return self.gemma3_12b.load_sd(sd)
@@ -195,14 +152,3 @@ def ltxav_te(dtype_llama=None, llama_quantization_metadata=None):
                 dtype = dtype_llama
             super().__init__(dtype_llama=dtype_llama, device=device, dtype=dtype, model_options=model_options)
     return LTXAVTEModel_
-
-def gemma3_te(dtype_llama=None, llama_quantization_metadata=None):
-    class Gemma3_12BModel_(Gemma3_12BModel):
-        def __init__(self, device="cpu", dtype=None, model_options={}):
-            if llama_quantization_metadata is not None:
-                model_options = model_options.copy()
-                model_options["llama_quantization_metadata"] = llama_quantization_metadata
-            if dtype_llama is not None:
-                dtype = dtype_llama
-            super().__init__(device=device, dtype=dtype, model_options=model_options)
-    return Gemma3_12BModel_

comfy/text_encoders/lumina2.py

@@ -1,23 +1,23 @@
 from comfy import sd1_clip
 from .spiece_tokenizer import SPieceTokenizer
 import comfy.text_encoders.llama
-from comfy.text_encoders.lt import Gemma3_Tokenizer
-import comfy.utils
+
 
 class Gemma2BTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        special_tokens = {"<end_of_turn>": 107}
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=2304, embedding_key='gemma2_2b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, tokenizer_data=tokenizer_data)
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=2304, embedding_key='gemma2_2b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False}, tokenizer_data=tokenizer_data)
 
     def state_dict(self):
         return {"spiece_model": self.tokenizer.serialize_model()}
 
-class Gemma3_4BTokenizer(Gemma3_Tokenizer, sd1_clip.SDTokenizer):
+class Gemma3_4BTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        special_tokens = {"<image_soft_token>": 262144, "<end_of_turn>": 106}
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=2560, embedding_key='gemma3_4b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, disable_weights=True, tokenizer_data=tokenizer_data)
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=2560, embedding_key='gemma3_4b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False}, disable_weights=True, tokenizer_data=tokenizer_data)
+
+    def state_dict(self):
+        return {"spiece_model": self.tokenizer.serialize_model()}
 
 class LuminaTokenizer(sd1_clip.SD1Tokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
@@ -40,20 +40,6 @@ class Gemma3_4BModel(sd1_clip.SDClipModel):
 
         super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma3_4B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
-class Gemma3_4B_Vision_Model(sd1_clip.SDClipModel):
-    def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
-        llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
-        if llama_quantization_metadata is not None:
-            model_options = model_options.copy()
-            model_options["quantization_metadata"] = llama_quantization_metadata
-
-        super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma3_4B_Vision, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
-
-    def process_tokens(self, tokens, device):
-        embeds, _, _, embeds_info = super().process_tokens(tokens, device)
-        comfy.utils.normalize_image_embeddings(embeds, embeds_info, self.transformer.model.config.hidden_size ** 0.5)
-        return embeds
-
 class LuminaModel(sd1_clip.SD1ClipModel):
     def __init__(self, device="cpu", dtype=None, model_options={}, name="gemma2_2b", clip_model=Gemma2_2BModel):
         super().__init__(device=device, dtype=dtype, name=name, clip_model=clip_model, model_options=model_options)
@@ -64,8 +50,6 @@ def te(dtype_llama=None, llama_quantization_metadata=None, model_type="gemma2_2b
         model = Gemma2_2BModel
     elif model_type == "gemma3_4b":
         model = Gemma3_4BModel
-    elif model_type == "gemma3_4b_vision":
-        model = Gemma3_4B_Vision_Model
 
     class LuminaTEModel_(LuminaModel):
         def __init__(self, device="cpu", dtype=None, model_options={}):

comfy/text_encoders/spiece_tokenizer.py

@@ -6,10 +6,9 @@ class SPieceTokenizer:
     def from_pretrained(path, **kwargs):
         return SPieceTokenizer(path, **kwargs)
 
-    def __init__(self, tokenizer_path, add_bos=False, add_eos=True, special_tokens=None):
+    def __init__(self, tokenizer_path, add_bos=False, add_eos=True):
         self.add_bos = add_bos
         self.add_eos = add_eos
-        self.special_tokens = special_tokens
         import sentencepiece
         if torch.is_tensor(tokenizer_path):
             tokenizer_path = tokenizer_path.numpy().tobytes()
@@ -28,32 +27,8 @@ class SPieceTokenizer:
         return out
 
     def __call__(self, string):
-        if self.special_tokens is not None:
-            import re
-            special_tokens_pattern = '|'.join(re.escape(token) for token in self.special_tokens.keys())
-            if special_tokens_pattern and re.search(special_tokens_pattern, string):
-                parts = re.split(f'({special_tokens_pattern})', string)
-                result = []
-                for part in parts:
-                    if not part:
-                        continue
-                    if part in self.special_tokens:
-                        result.append(self.special_tokens[part])
-                    else:
-                        encoded = self.tokenizer.encode(part, add_bos=False, add_eos=False)
-                        result.extend(encoded)
-                return {"input_ids": result}
-
         out = self.tokenizer.encode(string)
         return {"input_ids": out}
 
-    def decode(self, token_ids, skip_special_tokens=False):
-
-        if skip_special_tokens and self.special_tokens:
-            special_token_ids = set(self.special_tokens.values())
-            token_ids = [tid for tid in token_ids if tid not in special_token_ids]
-
-        return self.tokenizer.decode(token_ids)
-
     def serialize_model(self):
         return torch.ByteTensor(list(self.tokenizer.serialized_model_proto()))

comfy/utils.py

@@ -1418,11 +1418,3 @@ def deepcopy_list_dict(obj, memo=None):
 
     memo[obj_id] = res
     return res
-
-def normalize_image_embeddings(embeds, embeds_info, scale_factor):
-    """Normalize image embeddings to match text embedding scale"""
-    for info in embeds_info:
-        if info.get("type") == "image":
-            start_idx = info["index"]
-            end_idx = start_idx + info["size"]
-            embeds[:, start_idx:end_idx, :] /= scale_factor

comfy_api/latest/_input_impl/video_types.py

@@ -444,7 +444,7 @@ class VideoFromComponents(VideoInput):
             output.mux(packet)
 
             if audio_stream and self.__components.audio:
-                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().contiguous().numpy(), format='fltp', layout=layout)
+                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().numpy(), format='fltp', layout=layout)
                 frame.sample_rate = audio_sample_rate
                 frame.pts = 0
                 output.mux(audio_stream.encode(frame))

comfy_api/latest/_io.py

@@ -73,7 +73,6 @@ class RemoteOptions:
 class NumberDisplay(str, Enum):
     number = "number"
     slider = "slider"
-    gradient_slider = "gradientslider"
 
 
 class ControlAfterGenerate(str, Enum):
@@ -297,15 +296,13 @@ class Float(ComfyTypeIO):
         '''Float input.'''
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
                     default: float=None, min: float=None, max: float=None, step: float=None, round: float=None,
-                    display_mode: NumberDisplay=None, gradient_stops: list[list[float]]=None,
-                    socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
+                    display_mode: NumberDisplay=None, socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, display_name, optional, tooltip, lazy, default, socketless, None, force_input, extra_dict, raw_link, advanced)
             self.min = min
             self.max = max
             self.step = step
             self.round = round
             self.display_mode = display_mode
-            self.gradient_stops = gradient_stops
             self.default: float
 
         def as_dict(self):
@@ -315,7 +312,6 @@ class Float(ComfyTypeIO):
                 "step": self.step,
                 "round": self.round,
                 "display": self.display_mode,
-                "gradient_stops": self.gradient_stops,
             })
 
 @comfytype(io_type="STRING")
@@ -1213,30 +1209,6 @@ class Color(ComfyTypeIO):
       def as_dict(self):
           return super().as_dict()
 
-@comfytype(io_type="BOUNDING_BOX")
-class BoundingBox(ComfyTypeIO):
-    class BoundingBoxDict(TypedDict):
-        x: int
-        y: int
-        width: int
-        height: int
-    Type = BoundingBoxDict
-
-    class Input(WidgetInput):
-        def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None,
-                     socketless: bool=True, default: dict=None, component: str=None):
-            super().__init__(id, display_name, optional, tooltip, None, default, socketless)
-            self.component = component
-            if default is None:
-                self.default = {"x": 0, "y": 0, "width": 512, "height": 512}
-
-        def as_dict(self):
-            d = super().as_dict()
-            if self.component:
-                d["component"] = self.component
-            return d
-
-
 DYNAMIC_INPUT_LOOKUP: dict[str, Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]] = {}
 def register_dynamic_input_func(io_type: str, func: Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]):
     DYNAMIC_INPUT_LOOKUP[io_type] = func
@@ -1343,7 +1315,6 @@ class NodeInfoV1:
     api_node: bool=None
     price_badge: dict | None = None
     search_aliases: list[str]=None
-    essentials_category: str=None
 
 
 @dataclass
@@ -1465,8 +1436,6 @@ class Schema:
     """Flags a node as expandable, allowing NodeOutput to include 'expand' property."""
     accept_all_inputs: bool=False
     """When True, all inputs from the prompt will be passed to the node as kwargs, even if not defined in the schema."""
-    essentials_category: str | None = None
-    """Optional category for the Essentials tab. Path-based like category field (e.g., 'Basic', 'Image Tools/Editing')."""
 
     def validate(self):
         '''Validate the schema:
@@ -1573,7 +1542,6 @@ class Schema:
             python_module=getattr(cls, "RELATIVE_PYTHON_MODULE", "nodes"),
             price_badge=self.price_badge.as_dict(self.inputs) if self.price_badge is not None else None,
             search_aliases=self.search_aliases if self.search_aliases else None,
-            essentials_category=self.essentials_category,
         )
         return info
 
@@ -2222,6 +2190,5 @@ __all__ = [
     "ImageCompare",
     "PriceBadgeDepends",
     "PriceBadge",
-    "BoundingBox",
     "NodeReplace",
 ]

comfy_api_nodes/apis/elevenlabs.py

@@ -1,88 +0,0 @@
-from pydantic import BaseModel, Field
-
-
-class SpeechToTextRequest(BaseModel):
-    model_id: str = Field(...)
-    cloud_storage_url: str = Field(...)
-    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
-    tag_audio_events: bool | None = Field(None, description="Annotate sounds like (laughter) in transcript")
-    num_speakers: int | None = Field(None, description="Max speakers predicted")
-    timestamps_granularity: str = Field(default="word", description="Timing precision: none, word, or character")
-    diarize: bool | None = Field(None, description="Annotate which speaker is talking")
-    diarization_threshold: float | None = Field(None, description="Speaker separation sensitivity")
-    temperature: float | None = Field(None, description="Randomness control")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-
-
-class SpeechToTextWord(BaseModel):
-    text: str = Field(..., description="The word text")
-    type: str = Field(default="word", description="Type of text element (word, spacing, etc.)")
-    start: float | None = Field(None, description="Start time in seconds (when timestamps enabled)")
-    end: float | None = Field(None, description="End time in seconds (when timestamps enabled)")
-    speaker_id: str | None = Field(None, description="Speaker identifier when diarization is enabled")
-    logprob: float | None = Field(None, description="Log probability of the word")
-
-
-class SpeechToTextResponse(BaseModel):
-    language_code: str = Field(..., description="Detected or specified language code")
-    language_probability: float | None = Field(None, description="Confidence of language detection")
-    text: str = Field(..., description="Full transcript text")
-    words: list[SpeechToTextWord] | None = Field(None, description="Word-level timing information")
-
-
-class TextToSpeechVoiceSettings(BaseModel):
-    stability: float | None = Field(None, description="Voice stability")
-    similarity_boost: float | None = Field(None, description="Similarity boost")
-    style: float | None = Field(None, description="Style exaggeration")
-    use_speaker_boost: bool | None = Field(None, description="Boost similarity to original speaker")
-    speed: float | None = Field(None, description="Speech speed")
-
-
-class TextToSpeechRequest(BaseModel):
-    text: str = Field(..., description="Text to convert to speech")
-    model_id: str = Field(..., description="Model ID for TTS")
-    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
-    voice_settings: TextToSpeechVoiceSettings | None = Field(None, description="Voice settings")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")
-
-
-class TextToSoundEffectsRequest(BaseModel):
-    text: str = Field(..., description="Text prompt to convert into a sound effect")
-    duration_seconds: float = Field(..., description="Duration of generated sound in seconds")
-    prompt_influence: float = Field(..., description="How closely generation follows the prompt")
-    loop: bool | None = Field(None, description="Whether to create a smoothly looping sound effect")
-
-
-class AddVoiceRequest(BaseModel):
-    name: str = Field(..., description="Name that identifies the voice")
-    remove_background_noise: bool = Field(..., description="Remove background noise from voice samples")
-
-
-class AddVoiceResponse(BaseModel):
-    voice_id: str = Field(..., description="The newly created voice's unique identifier")
-
-
-class SpeechToSpeechRequest(BaseModel):
-    model_id: str = Field(..., description="Model ID for speech-to-speech")
-    voice_settings: str = Field(..., description="JSON string of voice settings")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-    remove_background_noise: bool = Field(..., description="Remove background noise from input audio")
-
-
-class DialogueInput(BaseModel):
-    text: str = Field(..., description="Text content to convert to speech")
-    voice_id: str = Field(..., description="Voice identifier for this dialogue segment")
-
-
-class DialogueSettings(BaseModel):
-    stability: float | None = Field(None, description="Voice stability (0-1)")
-
-
-class TextToDialogueRequest(BaseModel):
-    inputs: list[DialogueInput] = Field(..., description="List of dialogue segments")
-    model_id: str = Field(..., description="Model ID for dialogue generation")
-    language_code: str | None = Field(None, description="ISO-639-1 language code")
-    settings: DialogueSettings | None = Field(None, description="Voice settings")
-    seed: int | None = Field(None, description="Seed for deterministic sampling")
-    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")

comfy_api_nodes/apis/gemini.py

@@ -116,15 +116,9 @@ class GeminiGenerationConfig(BaseModel):
     topP: float | None = Field(None, ge=0.0, le=1.0)
 
 
-class GeminiImageOutputOptions(BaseModel):
-    mimeType: str = Field("image/png")
-    compressionQuality: int | None = Field(None)
-
-
 class GeminiImageConfig(BaseModel):
     aspectRatio: str | None = Field(None)
     imageSize: str | None = Field(None)
-    imageOutputOptions: GeminiImageOutputOptions = Field(default_factory=GeminiImageOutputOptions)
 
 
 class GeminiImageGenerationConfig(GeminiGenerationConfig):

comfy_api_nodes/apis/recraft.py

@@ -198,6 +198,11 @@ dict_recraft_substyles_v3 = {
 }
 
 
+class RecraftModel(str, Enum):
+    recraftv3 = 'recraftv3'
+    recraftv2 = 'recraftv2'
+
+
 class RecraftImageSize(str, Enum):
     res_1024x1024 = '1024x1024'
     res_1365x1024 = '1365x1024'
@@ -216,41 +221,6 @@ class RecraftImageSize(str, Enum):
     res_1707x1024 = '1707x1024'
 
 
-RECRAFT_V4_SIZES = [
-    "1024x1024",
-    "1536x768",
-    "768x1536",
-    "1280x832",
-    "832x1280",
-    "1216x896",
-    "896x1216",
-    "1152x896",
-    "896x1152",
-    "832x1344",
-    "1280x896",
-    "896x1280",
-    "1344x768",
-    "768x1344",
-]
-
-RECRAFT_V4_PRO_SIZES = [
-    "2048x2048",
-    "3072x1536",
-    "1536x3072",
-    "2560x1664",
-    "1664x2560",
-    "2432x1792",
-    "1792x2432",
-    "2304x1792",
-    "1792x2304",
-    "1664x2688",
-    "1434x1024",
-    "1024x1434",
-    "2560x1792",
-    "1792x2560",
-]
-
-
 class RecraftColorObject(BaseModel):
     rgb: list[int] = Field(..., description='An array of 3 integer values in range of 0...255 defining RGB Color Model')
 
@@ -264,16 +234,17 @@ class RecraftControlsObject(BaseModel):
 
 class RecraftImageGenerationRequest(BaseModel):
     prompt: str = Field(..., description='The text prompt describing the image to generate')
-    size: str | None = Field(None, description='The size of the generated image (e.g., "1024x1024")')
+    size: RecraftImageSize | None = Field(None, description='The size of the generated image (e.g., "1024x1024")')
     n: int = Field(..., description='The number of images to generate')
     negative_prompt: str | None = Field(None, description='A text description of undesired elements on an image')
-    model: str = Field(...)
+    model: RecraftModel | None = Field(RecraftModel.recraftv3, description='The model to use for generation (e.g., "recraftv3")')
     style: str | None = Field(None, description='The style to apply to the generated image (e.g., "digital_illustration")')
     substyle: str | None = Field(None, description='The substyle to apply to the generated image, depending on the style input')
     controls: RecraftControlsObject | None = Field(None, description='A set of custom parameters to tweak generation process')
     style_id: str | None = Field(None, description='Use a previously uploaded style as a reference; UUID')
     strength: float | None = Field(None, description='Defines the difference with the original image, should lie in [0, 1], where 0 means almost identical, and 1 means miserable similarity')
     random_seed: int | None = Field(None, description="Seed for video generation")
+    # text_layout
 
 
 class RecraftReturnedObject(BaseModel):

comfy_api_nodes/nodes_bfl.py

@@ -57,7 +57,6 @@ class FluxProUltraImageNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -201,7 +200,6 @@ class FluxKontextProImageNode(IO.ComfyNode):
                     "prompt_upsampling",
                     default=False,
                     tooltip="Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation, but results are nondeterministic (same seed will not produce exactly the same result).",
-                    advanced=True,
                 ),
                 IO.Image.Input(
                     "input_image",
@@ -298,7 +296,6 @@ class FluxProExpandNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "top",
@@ -436,7 +433,6 @@ class FluxProFillNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "guidance",
@@ -581,7 +577,6 @@ class Flux2ProImageNode(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation.",
-                    advanced=True,
                 ),
                 IO.Image.Input("images", optional=True, tooltip="Up to 9 images to be used as references."),
             ],

comfy_api_nodes/nodes_bytedance.py

@@ -114,7 +114,6 @@ class ByteDanceImageNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the image',
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -260,14 +259,12 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the image.',
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "fail_on_partial",
                     default=True,
                     tooltip="If enabled, abort execution if any requested images are missing or return an error.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -435,21 +432,18 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -567,21 +561,18 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -703,21 +694,18 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -846,7 +834,6 @@ class ByteDanceImageReferenceNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_elevenlabs.py

@@ -1,924 +0,0 @@
-import json
-import uuid
-
-from typing_extensions import override
-
-from comfy_api.latest import IO, ComfyExtension, Input
-from comfy_api_nodes.apis.elevenlabs import (
-    AddVoiceRequest,
-    AddVoiceResponse,
-    DialogueInput,
-    DialogueSettings,
-    SpeechToSpeechRequest,
-    SpeechToTextRequest,
-    SpeechToTextResponse,
-    TextToDialogueRequest,
-    TextToSoundEffectsRequest,
-    TextToSpeechRequest,
-    TextToSpeechVoiceSettings,
-)
-from comfy_api_nodes.util import (
-    ApiEndpoint,
-    audio_bytes_to_audio_input,
-    audio_ndarray_to_bytesio,
-    audio_tensor_to_contiguous_ndarray,
-    sync_op,
-    sync_op_raw,
-    upload_audio_to_comfyapi,
-    validate_string,
-)
-
-ELEVENLABS_MUSIC_SECTIONS = "ELEVENLABS_MUSIC_SECTIONS"  # Custom type for music sections
-ELEVENLABS_COMPOSITION_PLAN = "ELEVENLABS_COMPOSITION_PLAN"  # Custom type for composition plan
-ELEVENLABS_VOICE = "ELEVENLABS_VOICE"  # Custom type for voice selection
-
-# Predefined ElevenLabs voices: (voice_id, display_name, gender, accent)
-ELEVENLABS_VOICES = [
-    ("CwhRBWXzGAHq8TQ4Fs17", "Roger", "male", "american"),
-    ("EXAVITQu4vr4xnSDxMaL", "Sarah", "female", "american"),
-    ("FGY2WhTYpPnrIDTdsKH5", "Laura", "female", "american"),
-    ("IKne3meq5aSn9XLyUdCD", "Charlie", "male", "australian"),
-    ("JBFqnCBsd6RMkjVDRZzb", "George", "male", "british"),
-    ("N2lVS1w4EtoT3dr4eOWO", "Callum", "male", "american"),
-    ("SAz9YHcvj6GT2YYXdXww", "River", "neutral", "american"),
-    ("SOYHLrjzK2X1ezoPC6cr", "Harry", "male", "american"),
-    ("TX3LPaxmHKxFdv7VOQHJ", "Liam", "male", "american"),
-    ("Xb7hH8MSUJpSbSDYk0k2", "Alice", "female", "british"),
-    ("XrExE9yKIg1WjnnlVkGX", "Matilda", "female", "american"),
-    ("bIHbv24MWmeRgasZH58o", "Will", "male", "american"),
-    ("cgSgspJ2msm6clMCkdW9", "Jessica", "female", "american"),
-    ("cjVigY5qzO86Huf0OWal", "Eric", "male", "american"),
-    ("hpp4J3VqNfWAUOO0d1Us", "Bella", "female", "american"),
-    ("iP95p4xoKVk53GoZ742B", "Chris", "male", "american"),
-    ("nPczCjzI2devNBz1zQrb", "Brian", "male", "american"),
-    ("onwK4e9ZLuTAKqWW03F9", "Daniel", "male", "british"),
-    ("pFZP5JQG7iQjIQuC4Bku", "Lily", "female", "british"),
-    ("pNInz6obpgDQGcFmaJgB", "Adam", "male", "american"),
-    ("pqHfZKP75CvOlQylNhV4", "Bill", "male", "american"),
-]
-
-ELEVENLABS_VOICE_OPTIONS = [f"{name} ({gender}, {accent})" for _, name, gender, accent in ELEVENLABS_VOICES]
-ELEVENLABS_VOICE_MAP = {
-    f"{name} ({gender}, {accent})": voice_id for voice_id, name, gender, accent in ELEVENLABS_VOICES
-}
-
-
-class ElevenLabsSpeechToText(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsSpeechToText",
-            display_name="ElevenLabs Speech to Text",
-            category="api node/audio/ElevenLabs",
-            description="Transcribe audio to text. "
-            "Supports automatic language detection, speaker diarization, and audio event tagging.",
-            inputs=[
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Audio to transcribe.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "scribe_v2",
-                            [
-                                IO.Boolean.Input(
-                                    "tag_audio_events",
-                                    default=False,
-                                    tooltip="Annotate sounds like (laughter), (music), etc. in transcript.",
-                                ),
-                                IO.Boolean.Input(
-                                    "diarize",
-                                    default=False,
-                                    tooltip="Annotate which speaker is talking.",
-                                ),
-                                IO.Float.Input(
-                                    "diarization_threshold",
-                                    default=0.22,
-                                    min=0.1,
-                                    max=0.4,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speaker separation sensitivity. "
-                                    "Lower values are more sensitive to speaker changes.",
-                                ),
-                                IO.Float.Input(
-                                    "temperature",
-                                    default=0.0,
-                                    min=0.0,
-                                    max=2.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Randomness control. "
-                                    "0.0 uses model default. Higher values increase randomness.",
-                                ),
-                                IO.Combo.Input(
-                                    "timestamps_granularity",
-                                    options=["word", "character", "none"],
-                                    default="word",
-                                    tooltip="Timing precision for transcript words.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for transcription.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "num_speakers",
-                    default=0,
-                    min=0,
-                    max=32,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Maximum number of speakers to predict. Set to 0 for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=2147483647,
-                    tooltip="Seed for reproducibility (determinism not guaranteed).",
-                ),
-            ],
-            outputs=[
-                IO.String.Output(display_name="text"),
-                IO.String.Output(display_name="language_code"),
-                IO.String.Output(display_name="words_json"),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.0073,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        audio: Input.Audio,
-        model: dict,
-        language_code: str,
-        num_speakers: int,
-        seed: int,
-    ) -> IO.NodeOutput:
-        if model["diarize"] and num_speakers:
-            raise ValueError(
-                "Number of speakers cannot be specified when diarization is enabled. "
-                "Either disable diarization or set num_speakers to 0."
-            )
-        request = SpeechToTextRequest(
-            model_id=model["model"],
-            cloud_storage_url=await upload_audio_to_comfyapi(
-                cls, audio, container_format="mp4", codec_name="aac", mime_type="audio/mp4"
-            ),
-            language_code=language_code if language_code.strip() else None,
-            tag_audio_events=model["tag_audio_events"],
-            num_speakers=num_speakers if num_speakers > 0 else None,
-            timestamps_granularity=model["timestamps_granularity"],
-            diarize=model["diarize"],
-            diarization_threshold=model["diarization_threshold"] if model["diarize"] else None,
-            seed=seed,
-            temperature=model["temperature"],
-        )
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/speech-to-text", method="POST"),
-            response_model=SpeechToTextResponse,
-            data=request,
-            content_type="multipart/form-data",
-        )
-        words_json = json.dumps(
-            [w.model_dump(exclude_none=True) for w in response.words] if response.words else [],
-            indent=2,
-        )
-        return IO.NodeOutput(response.text, response.language_code, words_json)
-
-
-class ElevenLabsVoiceSelector(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsVoiceSelector",
-            display_name="ElevenLabs Voice Selector",
-            category="api node/audio/ElevenLabs",
-            description="Select a predefined ElevenLabs voice for text-to-speech generation.",
-            inputs=[
-                IO.Combo.Input(
-                    "voice",
-                    options=ELEVENLABS_VOICE_OPTIONS,
-                    tooltip="Choose a voice from the predefined ElevenLabs voices.",
-                ),
-            ],
-            outputs=[
-                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
-            ],
-            is_api_node=False,
-        )
-
-    @classmethod
-    def execute(cls, voice: str) -> IO.NodeOutput:
-        voice_id = ELEVENLABS_VOICE_MAP.get(voice)
-        if not voice_id:
-            raise ValueError(f"Unknown voice: {voice}")
-        return IO.NodeOutput(voice_id)
-
-
-class ElevenLabsTextToSpeech(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToSpeech",
-            display_name="ElevenLabs Text to Speech",
-            category="api node/audio/ElevenLabs",
-            description="Convert text to speech.",
-            inputs=[
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    "voice",
-                    tooltip="Voice to use for speech synthesis. Connect from Voice Selector or Instant Voice Clone.",
-                ),
-                IO.String.Input(
-                    "text",
-                    multiline=True,
-                    default="",
-                    tooltip="The text to convert to speech.",
-                ),
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.01,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.Combo.Input(
-                    "apply_text_normalization",
-                    options=["auto", "on", "off"],
-                    tooltip="Text normalization mode. 'auto' lets the system decide, "
-                    "'on' always applies normalization, 'off' skips it.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_multilingual_v2",
-                            [
-                                IO.Float.Input(
-                                    "speed",
-                                    default=1.0,
-                                    min=0.7,
-                                    max=1.3,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-                                ),
-                                IO.Float.Input(
-                                    "similarity_boost",
-                                    default=0.75,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-                                ),
-                                IO.Boolean.Input(
-                                    "use_speaker_boost",
-                                    default=False,
-                                    tooltip="Boost similarity to the original speaker voice.",
-                                ),
-                                IO.Float.Input(
-                                    "style",
-                                    default=0.0,
-                                    min=0.0,
-                                    max=0.2,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Style exaggeration. Higher values increase stylistic expression "
-                                    "but may reduce stability.",
-                                ),
-                            ],
-                        ),
-                        IO.DynamicCombo.Option(
-                            "eleven_v3",
-                            [
-                                IO.Float.Input(
-                                    "speed",
-                                    default=1.0,
-                                    min=0.7,
-                                    max=1.3,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-                                ),
-                                IO.Float.Input(
-                                    "similarity_boost",
-                                    default=0.75,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for text-to-speech.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=2147483647,
-                    tooltip="Seed for reproducibility (determinism not guaranteed).",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        voice: str,
-        text: str,
-        stability: float,
-        apply_text_normalization: str,
-        model: dict,
-        language_code: str,
-        seed: int,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        validate_string(text, min_length=1)
-        request = TextToSpeechRequest(
-            text=text,
-            model_id=model["model"],
-            language_code=language_code if language_code.strip() else None,
-            voice_settings=TextToSpeechVoiceSettings(
-                stability=stability,
-                similarity_boost=model["similarity_boost"],
-                speed=model["speed"],
-                use_speaker_boost=model.get("use_speaker_boost", None),
-                style=model.get("style", None),
-            ),
-            seed=seed,
-            apply_text_normalization=apply_text_normalization,
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path=f"/proxy/elevenlabs/v1/text-to-speech/{voice}",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=request,
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsAudioIsolation(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsAudioIsolation",
-            display_name="ElevenLabs Voice Isolation",
-            category="api node/audio/ElevenLabs",
-            description="Remove background noise from audio, isolating vocals or speech.",
-            inputs=[
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Audio to process for background noise removal.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        audio: Input.Audio,
-    ) -> IO.NodeOutput:
-        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
-        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/audio-isolation", method="POST"),
-            files={"audio": ("audio.mp4", audio_bytes_io, "audio/mp4")},
-            content_type="multipart/form-data",
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsTextToSoundEffects(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToSoundEffects",
-            display_name="ElevenLabs Text to Sound Effects",
-            category="api node/audio/ElevenLabs",
-            description="Generate sound effects from text descriptions.",
-            inputs=[
-                IO.String.Input(
-                    "text",
-                    multiline=True,
-                    default="",
-                    tooltip="Text description of the sound effect to generate.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_sfx_v2",
-                            [
-                                IO.Float.Input(
-                                    "duration",
-                                    default=5.0,
-                                    min=0.5,
-                                    max=30.0,
-                                    step=0.1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of generated sound in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "loop",
-                                    default=False,
-                                    tooltip="Create a smoothly looping sound effect.",
-                                ),
-                                IO.Float.Input(
-                                    "prompt_influence",
-                                    default=0.3,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="How closely generation follows the prompt. "
-                                    "Higher values make the sound follow the text more closely.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for sound effect generation.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.14,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        text: str,
-        model: dict,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        validate_string(text, min_length=1)
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path="/proxy/elevenlabs/v1/sound-generation",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=TextToSoundEffectsRequest(
-                text=text,
-                duration_seconds=model["duration"],
-                prompt_influence=model["prompt_influence"],
-                loop=model.get("loop", None),
-            ),
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsInstantVoiceClone(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsInstantVoiceClone",
-            display_name="ElevenLabs Instant Voice Clone",
-            category="api node/audio/ElevenLabs",
-            description="Create a cloned voice from audio samples. "
-            "Provide 1-8 audio recordings of the voice to clone.",
-            inputs=[
-                IO.Autogrow.Input(
-                    "files",
-                    template=IO.Autogrow.TemplatePrefix(
-                        IO.Audio.Input("audio"),
-                        prefix="audio",
-                        min=1,
-                        max=8,
-                    ),
-                    tooltip="Audio recordings for voice cloning.",
-                ),
-                IO.Boolean.Input(
-                    "remove_background_noise",
-                    default=False,
-                    tooltip="Remove background noise from voice samples using audio isolation.",
-                ),
-            ],
-            outputs=[
-                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(expr="""{"type":"usd","usd":0.15}"""),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        files: IO.Autogrow.Type,
-        remove_background_noise: bool,
-    ) -> IO.NodeOutput:
-        file_tuples: list[tuple[str, tuple[str, bytes, str]]] = []
-        for key in files:
-            audio = files[key]
-            sample_rate: int = audio["sample_rate"]
-            waveform = audio["waveform"]
-            audio_data_np = audio_tensor_to_contiguous_ndarray(waveform)
-            audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, sample_rate, "mp4", "aac")
-            file_tuples.append(("files", (f"{key}.mp4", audio_bytes_io.getvalue(), "audio/mp4")))
-
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/voices/add", method="POST"),
-            response_model=AddVoiceResponse,
-            data=AddVoiceRequest(
-                name=str(uuid.uuid4()),
-                remove_background_noise=remove_background_noise,
-            ),
-            files=file_tuples,
-            content_type="multipart/form-data",
-        )
-        return IO.NodeOutput(response.voice_id)
-
-
-ELEVENLABS_STS_VOICE_SETTINGS = [
-    IO.Float.Input(
-        "speed",
-        default=1.0,
-        min=0.7,
-        max=1.3,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-    ),
-    IO.Float.Input(
-        "similarity_boost",
-        default=0.75,
-        min=0.0,
-        max=1.0,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-    ),
-    IO.Boolean.Input(
-        "use_speaker_boost",
-        default=False,
-        tooltip="Boost similarity to the original speaker voice.",
-    ),
-    IO.Float.Input(
-        "style",
-        default=0.0,
-        min=0.0,
-        max=0.2,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Style exaggeration. Higher values increase stylistic expression but may reduce stability.",
-    ),
-]
-
-
-class ElevenLabsSpeechToSpeech(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsSpeechToSpeech",
-            display_name="ElevenLabs Speech to Speech",
-            category="api node/audio/ElevenLabs",
-            description="Transform speech from one voice to another while preserving the original content and emotion.",
-            inputs=[
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    "voice",
-                    tooltip="Target voice for the transformation. "
-                    "Connect from Voice Selector or Instant Voice Clone.",
-                ),
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Source audio to transform.",
-                ),
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.01,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_multilingual_sts_v2",
-                            ELEVENLABS_STS_VOICE_SETTINGS,
-                        ),
-                        IO.DynamicCombo.Option(
-                            "eleven_english_sts_v2",
-                            ELEVENLABS_STS_VOICE_SETTINGS,
-                        ),
-                    ],
-                    tooltip="Model to use for speech-to-speech transformation.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=0,
-                    min=0,
-                    max=4294967295,
-                    tooltip="Seed for reproducibility.",
-                ),
-                IO.Boolean.Input(
-                    "remove_background_noise",
-                    default=False,
-                    tooltip="Remove background noise from input audio using audio isolation.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        voice: str,
-        audio: Input.Audio,
-        stability: float,
-        model: dict,
-        output_format: str,
-        seed: int,
-        remove_background_noise: bool,
-    ) -> IO.NodeOutput:
-        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
-        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
-        voice_settings = TextToSpeechVoiceSettings(
-            stability=stability,
-            similarity_boost=model["similarity_boost"],
-            style=model["style"],
-            use_speaker_boost=model["use_speaker_boost"],
-            speed=model["speed"],
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path=f"/proxy/elevenlabs/v1/speech-to-speech/{voice}",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=SpeechToSpeechRequest(
-                model_id=model["model"],
-                voice_settings=voice_settings.model_dump_json(exclude_none=True),
-                seed=seed,
-                remove_background_noise=remove_background_noise,
-            ),
-            files={"audio": ("audio.mp4", audio_bytes_io.getvalue(), "audio/mp4")},
-            content_type="multipart/form-data",
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-def _generate_dialogue_inputs(count: int) -> list:
-    """Generate input widgets for a given number of dialogue entries."""
-    inputs = []
-    for i in range(1, count + 1):
-        inputs.extend(
-            [
-                IO.String.Input(
-                    f"text{i}",
-                    multiline=True,
-                    default="",
-                    tooltip=f"Text content for dialogue entry {i}.",
-                ),
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    f"voice{i}",
-                    tooltip=f"Voice for dialogue entry {i}. Connect from Voice Selector or Instant Voice Clone.",
-                ),
-            ]
-        )
-    return inputs
-
-
-class ElevenLabsTextToDialogue(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToDialogue",
-            display_name="ElevenLabs Text to Dialogue",
-            category="api node/audio/ElevenLabs",
-            description="Generate multi-speaker dialogue from text. Each dialogue entry has its own text and voice.",
-            inputs=[
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.5,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.Combo.Input(
-                    "apply_text_normalization",
-                    options=["auto", "on", "off"],
-                    tooltip="Text normalization mode. 'auto' lets the system decide, "
-                    "'on' always applies normalization, 'off' skips it.",
-                ),
-                IO.Combo.Input(
-                    "model",
-                    options=["eleven_v3"],
-                    tooltip="Model to use for dialogue generation.",
-                ),
-                IO.DynamicCombo.Input(
-                    "inputs",
-                    options=[
-                        IO.DynamicCombo.Option("1", _generate_dialogue_inputs(1)),
-                        IO.DynamicCombo.Option("2", _generate_dialogue_inputs(2)),
-                        IO.DynamicCombo.Option("3", _generate_dialogue_inputs(3)),
-                        IO.DynamicCombo.Option("4", _generate_dialogue_inputs(4)),
-                        IO.DynamicCombo.Option("5", _generate_dialogue_inputs(5)),
-                        IO.DynamicCombo.Option("6", _generate_dialogue_inputs(6)),
-                        IO.DynamicCombo.Option("7", _generate_dialogue_inputs(7)),
-                        IO.DynamicCombo.Option("8", _generate_dialogue_inputs(8)),
-                        IO.DynamicCombo.Option("9", _generate_dialogue_inputs(9)),
-                        IO.DynamicCombo.Option("10", _generate_dialogue_inputs(10)),
-                    ],
-                    tooltip="Number of dialogue entries.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=4294967295,
-                    tooltip="Seed for reproducibility.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        stability: float,
-        apply_text_normalization: str,
-        model: str,
-        inputs: dict,
-        language_code: str,
-        seed: int,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        num_entries = int(inputs["inputs"])
-        dialogue_inputs: list[DialogueInput] = []
-        for i in range(1, num_entries + 1):
-            text = inputs[f"text{i}"]
-            voice_id = inputs[f"voice{i}"]
-            validate_string(text, min_length=1)
-            dialogue_inputs.append(DialogueInput(text=text, voice_id=voice_id))
-        request = TextToDialogueRequest(
-            inputs=dialogue_inputs,
-            model_id=model,
-            language_code=language_code if language_code.strip() else None,
-            settings=DialogueSettings(stability=stability),
-            seed=seed,
-            apply_text_normalization=apply_text_normalization,
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path="/proxy/elevenlabs/v1/text-to-dialogue",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=request,
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsExtension(ComfyExtension):
-    @override
-    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
-        return [
-            ElevenLabsSpeechToText,
-            ElevenLabsVoiceSelector,
-            ElevenLabsTextToSpeech,
-            ElevenLabsAudioIsolation,
-            ElevenLabsTextToSoundEffects,
-            ElevenLabsInstantVoiceClone,
-            ElevenLabsSpeechToSpeech,
-            ElevenLabsTextToDialogue,
-        ]
-
-
-async def comfy_entrypoint() -> ElevenLabsExtension:
-    return ElevenLabsExtension()

comfy_api_nodes/nodes_gemini.py

@@ -6,7 +6,6 @@ See: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/infer
 import base64
 import os
 from enum import Enum
-from fnmatch import fnmatch
 from io import BytesIO
 from typing import Literal
 
@@ -120,13 +119,6 @@ async def create_image_parts(
     return image_parts
 
 
-def _mime_matches(mime: GeminiMimeType | None, pattern: str) -> bool:
-    """Check if a MIME type matches a pattern. Supports fnmatch globs (e.g. 'image/*')."""
-    if mime is None:
-        return False
-    return fnmatch(mime.value, pattern)
-
-
 def get_parts_by_type(response: GeminiGenerateContentResponse, part_type: Literal["text"] | str) -> list[GeminiPart]:
     """
     Filter response parts by their type.
@@ -159,9 +151,9 @@ def get_parts_by_type(response: GeminiGenerateContentResponse, part_type: Litera
         for part in candidate.content.parts:
             if part_type == "text" and part.text:
                 parts.append(part)
-            elif part.inlineData and _mime_matches(part.inlineData.mimeType, part_type):
+            elif part.inlineData and part.inlineData.mimeType == part_type:
                 parts.append(part)
-            elif part.fileData and _mime_matches(part.fileData.mimeType, part_type):
+            elif part.fileData and part.fileData.mimeType == part_type:
                 parts.append(part)
 
     if not parts and blocked_reasons:
@@ -186,7 +178,7 @@ def get_text_from_response(response: GeminiGenerateContentResponse) -> str:
 
 async def get_image_from_response(response: GeminiGenerateContentResponse) -> Input.Image:
     image_tensors: list[Input.Image] = []
-    parts = get_parts_by_type(response, "image/*")
+    parts = get_parts_by_type(response, "image/png")
     for part in parts:
         if part.inlineData:
             image_data = base64.b64decode(part.inlineData.data)
@@ -316,7 +308,6 @@ class GeminiNode(IO.ComfyNode):
                     default="",
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -594,7 +585,6 @@ class GeminiImage(IO.ComfyNode):
                     tooltip="Choose 'IMAGE' for image-only output, or "
                     "'IMAGE+TEXT' to return both the generated image and a text response.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.String.Input(
                     "system_prompt",
@@ -602,7 +592,6 @@ class GeminiImage(IO.ComfyNode):
                     default=GEMINI_IMAGE_SYS_PROMPT,
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -637,7 +626,7 @@ class GeminiImage(IO.ComfyNode):
 
         if not aspect_ratio:
             aspect_ratio = "auto"  # for backward compatability with old workflows; to-do remove this in December
-        image_config = GeminiImageConfig() if aspect_ratio == "auto" else GeminiImageConfig(aspectRatio=aspect_ratio)
+        image_config = GeminiImageConfig(aspectRatio=aspect_ratio)
 
         if images is not None:
             parts.extend(await create_image_parts(cls, images))
@@ -657,7 +646,7 @@ class GeminiImage(IO.ComfyNode):
                 ],
                 generationConfig=GeminiImageGenerationConfig(
                     responseModalities=(["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"]),
-                    imageConfig=image_config,
+                    imageConfig=None if aspect_ratio == "auto" else image_config,
                 ),
                 systemInstruction=gemini_system_prompt,
             ),
@@ -717,7 +706,6 @@ class GeminiImage2(IO.ComfyNode):
                     options=["IMAGE+TEXT", "IMAGE"],
                     tooltip="Choose 'IMAGE' for image-only output, or "
                     "'IMAGE+TEXT' to return both the generated image and a text response.",
-                    advanced=True,
                 ),
                 IO.Image.Input(
                     "images",
@@ -737,7 +725,6 @@ class GeminiImage2(IO.ComfyNode):
                     default=GEMINI_IMAGE_SYS_PROMPT,
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_hunyuan3d.py

@@ -54,7 +54,6 @@ class TencentTextToModelNode(IO.ComfyNode):
             node_id="TencentTextToModelNode",
             display_name="Hunyuan3D: Text to Model",
             category="api node/3d/Tencent",
-            essentials_category="3D",
             inputs=[
                 IO.Combo.Input(
                     "model",
@@ -169,7 +168,6 @@ class TencentImageToModelNode(IO.ComfyNode):
             node_id="TencentImageToModelNode",
             display_name="Hunyuan3D: Image(s) to Model",
             category="api node/3d/Tencent",
-            essentials_category="3D",
             inputs=[
                 IO.Combo.Input(
                     "model",

comfy_api_nodes/nodes_ideogram.py

@@ -261,7 +261,6 @@ class IdeogramV1(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -395,7 +394,6 @@ class IdeogramV2(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -413,7 +411,6 @@ class IdeogramV2(IO.ComfyNode):
                     default="NONE",
                     tooltip="Style type for generation (V2 only)",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.String.Input(
                     "negative_prompt",
@@ -567,7 +564,6 @@ class IdeogramV3(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -594,7 +590,6 @@ class IdeogramV3(IO.ComfyNode):
                     default="DEFAULT",
                     tooltip="Controls the trade-off between generation speed and quality",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Image.Input(
                     "character_image",

comfy_api_nodes/nodes_kling.py

@@ -2262,7 +2262,6 @@ class KlingLipSyncAudioToVideoNode(IO.ComfyNode):
             node_id="KlingLipSyncAudioToVideoNode",
             display_name="Kling Lip Sync Video with Audio",
             category="api node/video/Kling",
-            essentials_category="Video Generation",
             description="Kling Lip Sync Audio to Video Node. Syncs mouth movements in a video file to the audio content of an audio file. When using, ensure that the audio contains clearly distinguishable vocals and that the video contains a distinct face. The audio file should not be larger than 5MB. The video file should not be larger than 100MB, should have height/width between 720px and 1920px, and should be between 2s and 10s in length.",
             inputs=[
                 IO.Video.Input("video"),
@@ -2334,7 +2333,6 @@ class KlingLipSyncTextToVideoNode(IO.ComfyNode):
                     max=2.0,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Speech Rate. Valid range: 0.8~2.0, accurate to one decimal place.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -2456,7 +2454,6 @@ class KlingImageGenerationNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "image_type",
                     options=[i.value for i in KlingImageGenImageReferenceType],
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_fidelity",
@@ -2466,7 +2463,6 @@ class KlingImageGenerationNode(IO.ComfyNode):
                     step=0.01,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Reference intensity for user-uploaded images",
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "human_fidelity",
@@ -2476,7 +2472,6 @@ class KlingImageGenerationNode(IO.ComfyNode):
                     step=0.01,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Subject reference similarity",
-                    advanced=True,
                 ),
                 IO.Combo.Input("model_name", options=["kling-v3", "kling-v2", "kling-v1-5"]),
                 IO.Combo.Input(
@@ -2592,7 +2587,7 @@ class TextToVideoWithAudio(IO.ComfyNode):
                 IO.Combo.Input("mode", options=["pro"]),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
                 IO.Combo.Input("duration", options=[5, 10]),
-                IO.Boolean.Input("generate_audio", default=True, advanced=True),
+                IO.Boolean.Input("generate_audio", default=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -2660,7 +2655,7 @@ class ImageToVideoWithAudio(IO.ComfyNode):
                 IO.String.Input("prompt", multiline=True, tooltip="Positive text prompt."),
                 IO.Combo.Input("mode", options=["pro"]),
                 IO.Combo.Input("duration", options=[5, 10]),
-                IO.Boolean.Input("generate_audio", default=True, advanced=True),
+                IO.Boolean.Input("generate_audio", default=True),
             ],
             outputs=[
                 IO.Video.Output(),

comfy_api_nodes/nodes_ltxv.py

@@ -74,7 +74,6 @@ class TextToVideoNode(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="When true, the generated video will include AI-generated audio matching the scene.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -152,7 +151,6 @@ class ImageToVideoNode(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="When true, the generated video will include AI-generated audio matching the scene.",
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_magnific.py

@@ -110,13 +110,11 @@ class MagnificImageUpscalerCreativeNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "engine",
                     options=["automatic", "magnific_illusio", "magnific_sharpy", "magnific_sparkle"],
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "auto_downscale",
                     default=False,
                     tooltip="Automatically downscale input image if output would exceed maximum pixel limit.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -282,7 +280,6 @@ class MagnificImageUpscalerPreciseV2Node(IO.ComfyNode):
                     "auto_downscale",
                     default=False,
                     tooltip="Automatically downscale input image if output would exceed maximum resolution.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -443,7 +440,6 @@ class MagnificImageStyleTransferNode(IO.ComfyNode):
                         "softy",
                     ],
                     tooltip="Processing engine selection.",
-                    advanced=True,
                 ),
                 IO.DynamicCombo.Input(
                     "portrait_mode",
@@ -472,7 +468,6 @@ class MagnificImageStyleTransferNode(IO.ComfyNode):
                     default=True,
                     tooltip="When disabled, expect each generation to introduce a degree of randomness, "
                     "leading to more diverse outcomes.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -587,19 +582,16 @@ class MagnificImageRelightNode(IO.ComfyNode):
                     "interpolate_from_original",
                     default=False,
                     tooltip="Restricts generation freedom to match original more closely.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "change_background",
                     default=True,
                     tooltip="Modifies background based on prompt/reference.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "preserve_details",
                     default=True,
                     tooltip="Maintains texture and fine details from original.",
-                    advanced=True,
                 ),
                 IO.DynamicCombo.Input(
                     "advanced_settings",

comfy_api_nodes/nodes_meshy.py

@@ -58,12 +58,11 @@ class MeshyTextToModelNode(IO.ComfyNode):
                     ],
                     tooltip="When set to false, returns an unprocessed triangular mesh.",
                 ),
-                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"], advanced=True),
+                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"]),
                 IO.Combo.Input(
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -156,7 +155,6 @@ class MeshyRefineNode(IO.ComfyNode):
                     tooltip="Generate PBR Maps (metallic, roughness, normal) in addition to the base color. "
                     "Note: this should be set to false when using Sculpture style, "
                     "as Sculpture style generates its own set of PBR maps.",
-                    advanced=True,
                 ),
                 IO.String.Input(
                     "texture_prompt",
@@ -301,7 +299,6 @@ class MeshyImageToModelNode(IO.ComfyNode):
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -432,7 +429,7 @@ class MeshyMultiImageToModelNode(IO.ComfyNode):
                     ],
                     tooltip="When set to false, returns an unprocessed triangular mesh.",
                 ),
-                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"], advanced=True),
+                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"]),
                 IO.DynamicCombo.Input(
                     "should_texture",
                     options=[
@@ -469,7 +466,6 @@ class MeshyMultiImageToModelNode(IO.ComfyNode):
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -732,9 +728,8 @@ class MeshyTextureNode(IO.ComfyNode):
                     tooltip="Use the original UV of the model instead of generating new UVs. "
                     "When enabled, Meshy preserves existing textures from the uploaded model. "
                     "If the model has no original UV, the quality of the output might not be as good.",
-                    advanced=True,
                 ),
-                IO.Boolean.Input("pbr", default=False, advanced=True),
+                IO.Boolean.Input("pbr", default=False),
                 IO.String.Input(
                     "text_style_prompt",
                     default="",

comfy_api_nodes/nodes_openai.py

@@ -575,7 +575,6 @@ class OpenAIChatNode(IO.ComfyNode):
             node_id="OpenAIChatNode",
             display_name="OpenAI ChatGPT",
             category="api node/text/OpenAI",
-            essentials_category="Text Generation",
             description="Generate text responses from an OpenAI model.",
             inputs=[
                 IO.String.Input(
@@ -588,7 +587,6 @@ class OpenAIChatNode(IO.ComfyNode):
                     "persist_context",
                     default=False,
                     tooltip="This parameter is deprecated and has no effect.",
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "model",
@@ -858,7 +856,6 @@ class OpenAIChatConfig(IO.ComfyNode):
                     options=["auto", "disabled"],
                     default="auto",
                     tooltip="The truncation strategy to use for the model response. auto: If the context of this response and previous ones exceeds the model's context window size, the model will truncate the response to fit the context window by dropping input items in the middle of the conversation.disabled: If a model response will exceed the context window size for a model, the request will fail with a 400 error",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "max_output_tokens",
@@ -867,7 +864,6 @@ class OpenAIChatConfig(IO.ComfyNode):
                     max=16384,
                     tooltip="An upper bound for the number of tokens that can be generated for a response, including visible output tokens",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.String.Input(
                     "instructions",

comfy_api_nodes/nodes_recraft.py

@@ -1,4 +1,5 @@
 from io import BytesIO
+from typing import Optional, Union
 
 import aiohttp
 import torch
@@ -8,8 +9,6 @@ from typing_extensions import override
 from comfy.utils import ProgressBar
 from comfy_api.latest import IO, ComfyExtension
 from comfy_api_nodes.apis.recraft import (
-    RECRAFT_V4_PRO_SIZES,
-    RECRAFT_V4_SIZES,
     RecraftColor,
     RecraftColorChain,
     RecraftControls,
@@ -19,6 +18,7 @@ from comfy_api_nodes.apis.recraft import (
     RecraftImageGenerationResponse,
     RecraftImageSize,
     RecraftIO,
+    RecraftModel,
     RecraftStyle,
     RecraftStyleV3,
     get_v3_substyles,
@@ -39,7 +39,7 @@ async def handle_recraft_file_request(
     cls: type[IO.ComfyNode],
     image: torch.Tensor,
     path: str,
-    mask: torch.Tensor | None = None,
+    mask: Optional[torch.Tensor] = None,
     total_pixels: int = 4096 * 4096,
     timeout: int = 1024,
     request=None,
@@ -73,11 +73,11 @@ async def handle_recraft_file_request(
 def recraft_multipart_parser(
     data,
     parent_key=None,
-    formatter: type[callable] | None = None,
-    converted_to_check: list[list] | None = None,
+    formatter: Optional[type[callable]] = None,
+    converted_to_check: Optional[list[list]] = None,
     is_list: bool = False,
     return_mode: str = "formdata",  # "dict" | "formdata"
-) -> dict | aiohttp.FormData:
+) -> Union[dict, aiohttp.FormData]:
     """
     Formats data such that multipart/form-data will work with aiohttp library when both files and data are present.
 
@@ -309,7 +309,7 @@ class RecraftStyleInfiniteStyleLibrary(IO.ComfyNode):
             node_id="RecraftStyleV3InfiniteStyleLibrary",
             display_name="Recraft Style - Infinite Style Library",
             category="api node/image/Recraft",
-            description="Choose style based on preexisting UUID from Recraft's Infinite Style Library.",
+            description="Select style based on preexisting UUID from Recraft's Infinite Style Library.",
             inputs=[
                 IO.String.Input("style_id", default="", tooltip="UUID of style from Infinite Style Library."),
             ],
@@ -485,7 +485,7 @@ class RecraftTextToImageNode(IO.ComfyNode):
             data=RecraftImageGenerationRequest(
                 prompt=prompt,
                 negative_prompt=negative_prompt,
-                model="recraftv3",
+                model=RecraftModel.recraftv3,
                 size=size,
                 n=n,
                 style=recraft_style.style,
@@ -598,7 +598,7 @@ class RecraftImageToImageNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model="recraftv3",
+            model=RecraftModel.recraftv3,
             n=n,
             strength=round(strength, 2),
             style=recraft_style.style,
@@ -698,7 +698,7 @@ class RecraftImageInpaintingNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model="recraftv3",
+            model=RecraftModel.recraftv3,
             n=n,
             style=recraft_style.style,
             substyle=recraft_style.substyle,
@@ -810,7 +810,7 @@ class RecraftTextToVectorNode(IO.ComfyNode):
             data=RecraftImageGenerationRequest(
                 prompt=prompt,
                 negative_prompt=negative_prompt,
-                model="recraftv3",
+                model=RecraftModel.recraftv3,
                 size=size,
                 n=n,
                 style=recraft_style.style,
@@ -933,7 +933,7 @@ class RecraftReplaceBackgroundNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model="recraftv3",
+            model=RecraftModel.recraftv3,
             n=n,
             style=recraft_style.style,
             substyle=recraft_style.substyle,
@@ -963,7 +963,6 @@ class RecraftRemoveBackgroundNode(IO.ComfyNode):
             node_id="RecraftRemoveBackgroundNode",
             display_name="Recraft Remove Background",
             category="api node/image/Recraft",
-            essentials_category="Image Tools",
             description="Remove background from image, and return processed image and mask.",
             inputs=[
                 IO.Image.Input("image"),
@@ -1079,252 +1078,6 @@ class RecraftCreativeUpscaleNode(RecraftCrispUpscaleNode):
         )
 
 
-class RecraftV4TextToImageNode(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="RecraftV4TextToImageNode",
-            display_name="Recraft V4 Text to Image",
-            category="api node/image/Recraft",
-            description="Generates images using Recraft V4 or V4 Pro models.",
-            inputs=[
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                    tooltip="Prompt for the image generation. Maximum 10,000 characters.",
-                ),
-                IO.String.Input(
-                    "negative_prompt",
-                    multiline=True,
-                    tooltip="An optional text description of undesired elements on an image.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "recraftv4",
-                            [
-                                IO.Combo.Input(
-                                    "size",
-                                    options=RECRAFT_V4_SIZES,
-                                    default="1024x1024",
-                                    tooltip="The size of the generated image.",
-                                ),
-                            ],
-                        ),
-                        IO.DynamicCombo.Option(
-                            "recraftv4_pro",
-                            [
-                                IO.Combo.Input(
-                                    "size",
-                                    options=RECRAFT_V4_PRO_SIZES,
-                                    default="2048x2048",
-                                    tooltip="The size of the generated image.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="The model to use for generation.",
-                ),
-                IO.Int.Input(
-                    "n",
-                    default=1,
-                    min=1,
-                    max=6,
-                    tooltip="The number of images to generate.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=0,
-                    min=0,
-                    max=0xFFFFFFFFFFFFFFFF,
-                    control_after_generate=True,
-                    tooltip="Seed to determine if node should re-run; "
-                    "actual results are nondeterministic regardless of seed.",
-                ),
-                IO.Custom(RecraftIO.CONTROLS).Input(
-                    "recraft_controls",
-                    tooltip="Optional additional controls over the generation via the Recraft Controls node.",
-                    optional=True,
-                ),
-            ],
-            outputs=[
-                IO.Image.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model", "n"]),
-                expr="""
-                (
-                    $prices := {"recraftv4": 0.04, "recraftv4_pro": 0.25};
-                    {"type":"usd","usd": $lookup($prices, widgets.model) * widgets.n}
-                )
-                """,
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        prompt: str,
-        negative_prompt: str,
-        model: dict,
-        n: int,
-        seed: int,
-        recraft_controls: RecraftControls | None = None,
-    ) -> IO.NodeOutput:
-        validate_string(prompt, strip_whitespace=False, min_length=1, max_length=10000)
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/recraft/image_generation", method="POST"),
-            response_model=RecraftImageGenerationResponse,
-            data=RecraftImageGenerationRequest(
-                prompt=prompt,
-                negative_prompt=negative_prompt if negative_prompt else None,
-                model=model["model"],
-                size=model["size"],
-                n=n,
-                controls=recraft_controls.create_api_model() if recraft_controls else None,
-            ),
-            max_retries=1,
-        )
-        images = []
-        for data in response.data:
-            with handle_recraft_image_output():
-                image = bytesio_to_image_tensor(await download_url_as_bytesio(data.url, timeout=1024))
-            if len(image.shape) < 4:
-                image = image.unsqueeze(0)
-            images.append(image)
-        return IO.NodeOutput(torch.cat(images, dim=0))
-
-
-class RecraftV4TextToVectorNode(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="RecraftV4TextToVectorNode",
-            display_name="Recraft V4 Text to Vector",
-            category="api node/image/Recraft",
-            description="Generates SVG using Recraft V4 or V4 Pro models.",
-            inputs=[
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                    tooltip="Prompt for the image generation. Maximum 10,000 characters.",
-                ),
-                IO.String.Input(
-                    "negative_prompt",
-                    multiline=True,
-                    tooltip="An optional text description of undesired elements on an image.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "recraftv4",
-                            [
-                                IO.Combo.Input(
-                                    "size",
-                                    options=RECRAFT_V4_SIZES,
-                                    default="1024x1024",
-                                    tooltip="The size of the generated image.",
-                                ),
-                            ],
-                        ),
-                        IO.DynamicCombo.Option(
-                            "recraftv4_pro",
-                            [
-                                IO.Combo.Input(
-                                    "size",
-                                    options=RECRAFT_V4_PRO_SIZES,
-                                    default="2048x2048",
-                                    tooltip="The size of the generated image.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="The model to use for generation.",
-                ),
-                IO.Int.Input(
-                    "n",
-                    default=1,
-                    min=1,
-                    max=6,
-                    tooltip="The number of images to generate.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=0,
-                    min=0,
-                    max=0xFFFFFFFFFFFFFFFF,
-                    control_after_generate=True,
-                    tooltip="Seed to determine if node should re-run; "
-                    "actual results are nondeterministic regardless of seed.",
-                ),
-                IO.Custom(RecraftIO.CONTROLS).Input(
-                    "recraft_controls",
-                    tooltip="Optional additional controls over the generation via the Recraft Controls node.",
-                    optional=True,
-                ),
-            ],
-            outputs=[
-                IO.SVG.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model", "n"]),
-                expr="""
-                (
-                    $prices := {"recraftv4": 0.08, "recraftv4_pro": 0.30};
-                    {"type":"usd","usd": $lookup($prices, widgets.model) * widgets.n}
-                )
-                """,
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        prompt: str,
-        negative_prompt: str,
-        model: dict,
-        n: int,
-        seed: int,
-        recraft_controls: RecraftControls | None = None,
-    ) -> IO.NodeOutput:
-        validate_string(prompt, strip_whitespace=False, min_length=1, max_length=10000)
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/recraft/image_generation", method="POST"),
-            response_model=RecraftImageGenerationResponse,
-            data=RecraftImageGenerationRequest(
-                prompt=prompt,
-                negative_prompt=negative_prompt if negative_prompt else None,
-                model=model["model"],
-                size=model["size"],
-                n=n,
-                style="vector_illustration",
-                substyle=None,
-                controls=recraft_controls.create_api_model() if recraft_controls else None,
-            ),
-            max_retries=1,
-        )
-        svg_data = []
-        for data in response.data:
-            svg_data.append(await download_url_as_bytesio(data.url, timeout=1024))
-        return IO.NodeOutput(SVG(svg_data))
-
-
 class RecraftExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -1345,8 +1098,6 @@ class RecraftExtension(ComfyExtension):
             RecraftCreateStyleNode,
             RecraftColorRGBNode,
             RecraftControlsNode,
-            RecraftV4TextToImageNode,
-            RecraftV4TextToVectorNode,
         ]
 
 

comfy_api_nodes/nodes_rodin.py

@@ -493,7 +493,7 @@ class Rodin3D_Gen2(IO.ComfyNode):
                     default="500K-Triangle",
                     optional=True,
                 ),
-                IO.Boolean.Input("TAPose", default=False, advanced=True),
+                IO.Boolean.Input("TAPose", default=False),
             ],
             outputs=[
                 IO.String.Output(display_name="3D Model Path"),  # for backward compatibility only
@@ -505,9 +505,6 @@ class Rodin3D_Gen2(IO.ComfyNode):
                 IO.Hidden.unique_id,
             ],
             is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.4}""",
-            ),
         )
 
     @classmethod

comfy_api_nodes/nodes_stability.py

@@ -86,7 +86,6 @@ class StabilityStableImageUltraNode(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -108,7 +107,6 @@ class StabilityStableImageUltraNode(IO.ComfyNode):
                     tooltip="A blurb of text describing what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_denoise",
@@ -220,7 +218,6 @@ class StabilityStableImageSD_3_5Node(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "cfg_scale",
@@ -250,7 +247,6 @@ class StabilityStableImageSD_3_5Node(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_denoise",
@@ -388,7 +384,6 @@ class StabilityUpscaleConservativeNode(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -479,7 +474,6 @@ class StabilityUpscaleCreativeNode(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -497,7 +491,6 @@ class StabilityUpscaleCreativeNode(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -631,7 +624,6 @@ class StabilityTextToAudio(IO.ComfyNode):
             node_id="StabilityTextToAudio",
             display_name="Stability AI Text To Audio",
             category="api node/audio/Stability AI",
-            essentials_category="Audio",
             description=cleandoc(cls.__doc__ or ""),
             inputs=[
                 IO.Combo.Input(
@@ -667,7 +659,6 @@ class StabilityTextToAudio(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -745,7 +736,6 @@ class StabilityAudioToAudio(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "strength",
@@ -839,7 +829,6 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "mask_start",
@@ -848,7 +837,6 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     max=190,
                     step=1,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "mask_end",
@@ -857,7 +845,6 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     max=190,
                     step=1,
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_topaz.py

@@ -63,14 +63,12 @@ class TopazImageEnhance(IO.ComfyNode):
                     "subject_detection",
                     options=["All", "Foreground", "Background"],
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "face_enhancement",
                     default=True,
                     optional=True,
                     tooltip="Enhance faces (if present) during processing.",
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "face_enhancement_creativity",
@@ -81,7 +79,6 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Set the creativity level for face enhancement.",
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "face_enhancement_strength",
@@ -92,7 +89,6 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Controls how sharp enhanced faces are relative to the background.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "crop_to_fill",
@@ -100,7 +96,6 @@ class TopazImageEnhance(IO.ComfyNode):
                     optional=True,
                     tooltip="By default, the image is letterboxed when the output aspect ratio differs. "
                     "Enable to crop the image to fill the output dimensions.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "output_width",
@@ -111,7 +106,6 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Zero value means to calculate automatically (usually it will be original size or output_height if specified).",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "output_height",
@@ -122,7 +116,6 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Zero value means to output in the same height as original or output width.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "creativity",
@@ -138,14 +131,12 @@ class TopazImageEnhance(IO.ComfyNode):
                     default=True,
                     optional=True,
                     tooltip="Preserve subjects' facial identity.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "color_preservation",
                     default=True,
                     optional=True,
                     tooltip="Preserve the original colors.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -243,10 +234,9 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default="low",
                     tooltip="Creativity level (applies only to Starlight (Astra) Creative).",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input("interpolation_enabled", default=False, optional=True),
-                IO.Combo.Input("interpolation_model", options=["apo-8"], default="apo-8", optional=True, advanced=True),
+                IO.Combo.Input("interpolation_model", options=["apo-8"], default="apo-8", optional=True),
                 IO.Int.Input(
                     "interpolation_slowmo",
                     default=1,
@@ -256,7 +246,6 @@ class TopazVideoEnhance(IO.ComfyNode):
                     tooltip="Slow-motion factor applied to the input video. "
                     "For example, 2 makes the output twice as slow and doubles the duration.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "interpolation_frame_rate",
@@ -272,7 +261,6 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default=False,
                     tooltip="Analyze the input for duplicate frames and remove them.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Float.Input(
                     "interpolation_duplicate_threshold",
@@ -283,7 +271,6 @@ class TopazVideoEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     tooltip="Detection sensitivity for duplicate frames.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "dynamic_compression_level",
@@ -291,7 +278,6 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default="Low",
                     tooltip="CQP level.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_tripo.py

@@ -89,13 +89,13 @@ class TripoTextToModelNode(IO.ComfyNode):
                 IO.Combo.Input("style", options=TripoStyle, default="None", optional=True),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("image_seed", default=42, optional=True, advanced=True),
-                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
-                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
-                IO.Int.Input("face_limit", default=-1, min=-1, max=2000000, optional=True, advanced=True),
-                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("image_seed", default=42, optional=True),
+                IO.Int.Input("model_seed", default=42, optional=True),
+                IO.Int.Input("texture_seed", default=42, optional=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
+                IO.Int.Input("face_limit", default=-1, min=-1, max=2000000, optional=True),
+                IO.Boolean.Input("quad", default=False, optional=True),
+                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -210,18 +210,18 @@ class TripoImageToModelNode(IO.ComfyNode):
                 IO.Combo.Input("style", options=TripoStyle, default="None", optional=True),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
+                IO.Int.Input("model_seed", default=42, optional=True),
                 IO.Combo.Input(
-                    "orientation", options=TripoOrientation, default=TripoOrientation.DEFAULT, optional=True, advanced=True
+                    "orientation", options=TripoOrientation, default=TripoOrientation.DEFAULT, optional=True
                 ),
-                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("texture_seed", default=42, optional=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True, advanced=True
+                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
                 ),
-                IO.Int.Input("face_limit", default=-1, min=-1, max=500000, optional=True, advanced=True),
-                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("face_limit", default=-1, min=-1, max=500000, optional=True),
+                IO.Boolean.Input("quad", default=False, optional=True),
+                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -347,19 +347,18 @@ class TripoMultiviewToModelNode(IO.ComfyNode):
                     options=TripoOrientation,
                     default=TripoOrientation.DEFAULT,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
-                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("model_seed", default=42, optional=True),
+                IO.Int.Input("texture_seed", default=42, optional=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True, advanced=True
+                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
                 ),
-                IO.Int.Input("face_limit", default=-1, min=-1, max=500000, optional=True, advanced=True),
-                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("face_limit", default=-1, min=-1, max=500000, optional=True),
+                IO.Boolean.Input("quad", default=False, optional=True),
+                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -475,10 +474,10 @@ class TripoTextureNode(IO.ComfyNode):
                 IO.Custom("MODEL_TASK_ID").Input("model_task_id"),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("texture_seed", default=42, optional=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True, advanced=True
+                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
                 ),
             ],
             outputs=[
@@ -683,14 +682,13 @@ class TripoConversionNode(IO.ComfyNode):
             inputs=[
                 IO.Custom("MODEL_TASK_ID,RIG_TASK_ID,RETARGET_TASK_ID").Input("original_model_task_id"),
                 IO.Combo.Input("format", options=["GLTF", "USDZ", "FBX", "OBJ", "STL", "3MF"]),
-                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("quad", default=False, optional=True),
                 IO.Int.Input(
                     "face_limit",
                     default=-1,
                     min=-1,
                     max=2000000,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "texture_size",
@@ -698,53 +696,47 @@ class TripoConversionNode(IO.ComfyNode):
                     min=128,
                     max=4096,
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "texture_format",
                     options=["BMP", "DPX", "HDR", "JPEG", "OPEN_EXR", "PNG", "TARGA", "TIFF", "WEBP"],
                     default="JPEG",
                     optional=True,
-                    advanced=True,
                 ),
-                IO.Boolean.Input("force_symmetry", default=False, optional=True, advanced=True),
-                IO.Boolean.Input("flatten_bottom", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("force_symmetry", default=False, optional=True),
+                IO.Boolean.Input("flatten_bottom", default=False, optional=True),
                 IO.Float.Input(
                     "flatten_bottom_threshold",
                     default=0.0,
                     min=0.0,
                     max=1.0,
                     optional=True,
-                    advanced=True,
                 ),
-                IO.Boolean.Input("pivot_to_center_bottom", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("pivot_to_center_bottom", default=False, optional=True),
                 IO.Float.Input(
                     "scale_factor",
                     default=1.0,
                     min=0.0,
                     optional=True,
-                    advanced=True,
                 ),
-                IO.Boolean.Input("with_animation", default=False, optional=True, advanced=True),
-                IO.Boolean.Input("pack_uv", default=False, optional=True, advanced=True),
-                IO.Boolean.Input("bake", default=False, optional=True, advanced=True),
-                IO.String.Input("part_names", default="", optional=True, advanced=True),  # comma-separated list
+                IO.Boolean.Input("with_animation", default=False, optional=True),
+                IO.Boolean.Input("pack_uv", default=False, optional=True),
+                IO.Boolean.Input("bake", default=False, optional=True),
+                IO.String.Input("part_names", default="", optional=True),  # comma-separated list
                 IO.Combo.Input(
                     "fbx_preset",
                     options=["blender", "mixamo", "3dsmax"],
                     default="blender",
                     optional=True,
-                    advanced=True,
                 ),
-                IO.Boolean.Input("export_vertex_colors", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("export_vertex_colors", default=False, optional=True),
                 IO.Combo.Input(
                     "export_orientation",
                     options=["align_image", "default"],
                     default="default",
                     optional=True,
-                    advanced=True,
                 ),
-                IO.Boolean.Input("animate_in_place", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("animate_in_place", default=False, optional=True),
             ],
             outputs=[],
             hidden=[

comfy_api_nodes/nodes_veo2.py

@@ -81,7 +81,6 @@ class VeoVideoGenerationNode(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "person_generation",
@@ -89,7 +88,6 @@ class VeoVideoGenerationNode(IO.ComfyNode):
                     default="ALLOW",
                     tooltip="Whether to allow generating people in the video",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -301,7 +299,6 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
                     default=True,
                     tooltip="This parameter is deprecated and ignored.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "person_generation",
@@ -309,7 +306,6 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
                     default="ALLOW",
                     tooltip="Whether to allow generating people in the video",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",

comfy_api_nodes/nodes_vidu.py

@@ -54,7 +54,6 @@ async def execute_task(
         response_model=TaskStatusResponse,
         status_extractor=lambda r: r.state,
         progress_extractor=lambda r: r.progress,
-        price_extractor=lambda r: r.credits * 0.005 if r.credits is not None else None,
         max_poll_attempts=max_poll_attempts,
     )
     if not response.creations:
@@ -112,14 +111,12 @@ class ViduTextToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -210,14 +207,12 @@ class ViduImageToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -318,14 +313,12 @@ class ViduReferenceVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -432,14 +425,12 @@ class ViduStartEndToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -519,12 +510,11 @@ class Vidu2TextToVideoNode(IO.ComfyNode):
                     control_after_generate=True,
                 ),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "3:4", "4:3", "1:1"]),
-                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
+                IO.Combo.Input("resolution", options=["720p", "1080p"]),
                 IO.Boolean.Input(
                     "background_music",
                     default=False,
                     tooltip="Whether to add background music to the generated video.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -618,13 +608,11 @@ class Vidu2ImageToVideoNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "resolution",
                     options=["720p", "1080p"],
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -738,7 +726,6 @@ class Vidu2ReferenceVideoNode(IO.ComfyNode):
                     "audio",
                     default=False,
                     tooltip="When enabled video will contain generated speech and background music based on the prompt.",
-                    advanced=True,
                 ),
                 IO.Int.Input(
                     "duration",
@@ -758,12 +745,11 @@ class Vidu2ReferenceVideoNode(IO.ComfyNode):
                     control_after_generate=True,
                 ),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "4:3", "3:4", "1:1"]),
-                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
+                IO.Combo.Input("resolution", options=["720p", "1080p"]),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -877,12 +863,11 @@ class Vidu2StartEndToVideoNode(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     control_after_generate=True,
                 ),
-                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
+                IO.Combo.Input("resolution", options=["720p", "1080p"]),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -1321,36 +1306,6 @@ class Vidu3TextToVideoNode(IO.ComfyNode):
                                 ),
                             ],
                         ),
-                        IO.DynamicCombo.Option(
-                            "viduq3-turbo",
-                            [
-                                IO.Combo.Input(
-                                    "aspect_ratio",
-                                    options=["16:9", "9:16", "3:4", "4:3", "1:1"],
-                                    tooltip="The aspect ratio of the output video.",
-                                ),
-                                IO.Combo.Input(
-                                    "resolution",
-                                    options=["720p", "1080p"],
-                                    tooltip="Resolution of the output video.",
-                                ),
-                                IO.Int.Input(
-                                    "duration",
-                                    default=5,
-                                    min=1,
-                                    max=16,
-                                    step=1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of the output video in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "audio",
-                                    default=False,
-                                    tooltip="When enabled, outputs video with sound "
-                                    "(including dialogue and sound effects).",
-                                ),
-                            ],
-                        ),
                     ],
                     tooltip="Model to use for video generation.",
                 ),
@@ -1379,20 +1334,13 @@ class Vidu3TextToVideoNode(IO.ComfyNode):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model", "model.duration", "model.resolution"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model.duration", "model.resolution"]),
                 expr="""
                 (
                   $res := $lookup(widgets, "model.resolution");
-                  $d := $lookup(widgets, "model.duration");
-                  $contains(widgets.model, "turbo")
-                    ? (
-                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
-                    : (
-                        $rate := $lookup({"720p": 0.15, "1080p": 0.16}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
+                  $base := $lookup({"720p": 0.075, "1080p": 0.1}, $res);
+                  $perSec := $lookup({"720p": 0.025, "1080p": 0.05}, $res);
+                  {"type":"usd","usd": $base + $perSec * ($lookup(widgets, "model.duration") - 1)}
                 )
                 """,
             ),
@@ -1461,31 +1409,6 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
                                 ),
                             ],
                         ),
-                        IO.DynamicCombo.Option(
-                            "viduq3-turbo",
-                            [
-                                IO.Combo.Input(
-                                    "resolution",
-                                    options=["720p", "1080p"],
-                                    tooltip="Resolution of the output video.",
-                                ),
-                                IO.Int.Input(
-                                    "duration",
-                                    default=5,
-                                    min=1,
-                                    max=16,
-                                    step=1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of the output video in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "audio",
-                                    default=False,
-                                    tooltip="When enabled, outputs video with sound "
-                                    "(including dialogue and sound effects).",
-                                ),
-                            ],
-                        ),
                     ],
                     tooltip="Model to use for video generation.",
                 ),
@@ -1519,20 +1442,13 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model", "model.duration", "model.resolution"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model.duration", "model.resolution"]),
                 expr="""
                 (
                   $res := $lookup(widgets, "model.resolution");
-                  $d := $lookup(widgets, "model.duration");
-                  $contains(widgets.model, "turbo")
-                    ? (
-                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
-                    : (
-                        $rate := $lookup({"720p": 0.15, "1080p": 0.16, "2k": 0.2}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
+                  $base := $lookup({"720p": 0.075, "1080p": 0.275, "2k": 0.35}, $res);
+                  $perSec := $lookup({"720p": 0.05, "1080p": 0.075, "2k": 0.075}, $res);
+                  {"type":"usd","usd": $base + $perSec * ($lookup(widgets, "model.duration") - 1)}
                 )
                 """,
             ),
@@ -1565,145 +1481,6 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
         return IO.NodeOutput(await download_url_to_video_output(results[0].url))
 
 
-class Vidu3StartEndToVideoNode(IO.ComfyNode):
-
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="Vidu3StartEndToVideoNode",
-            display_name="Vidu Q3 Start/End Frame-to-Video Generation",
-            category="api node/video/Vidu",
-            description="Generate a video from a start frame, an end frame, and a prompt.",
-            inputs=[
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "viduq3-pro",
-                            [
-                                IO.Combo.Input(
-                                    "resolution",
-                                    options=["720p", "1080p"],
-                                    tooltip="Resolution of the output video.",
-                                ),
-                                IO.Int.Input(
-                                    "duration",
-                                    default=5,
-                                    min=1,
-                                    max=16,
-                                    step=1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of the output video in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "audio",
-                                    default=False,
-                                    tooltip="When enabled, outputs video with sound "
-                                    "(including dialogue and sound effects).",
-                                ),
-                            ],
-                        ),
-                        IO.DynamicCombo.Option(
-                            "viduq3-turbo",
-                            [
-                                IO.Combo.Input(
-                                    "resolution",
-                                    options=["720p", "1080p"],
-                                    tooltip="Resolution of the output video.",
-                                ),
-                                IO.Int.Input(
-                                    "duration",
-                                    default=5,
-                                    min=1,
-                                    max=16,
-                                    step=1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of the output video in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "audio",
-                                    default=False,
-                                    tooltip="When enabled, outputs video with sound "
-                                    "(including dialogue and sound effects).",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for video generation.",
-                ),
-                IO.Image.Input("first_frame"),
-                IO.Image.Input("end_frame"),
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                    tooltip="Prompt description (max 2000 characters).",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=2147483647,
-                    step=1,
-                    display_mode=IO.NumberDisplay.number,
-                    control_after_generate=True,
-                ),
-            ],
-            outputs=[
-                IO.Video.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model", "model.duration", "model.resolution"]),
-                expr="""
-                (
-                  $res := $lookup(widgets, "model.resolution");
-                  $d := $lookup(widgets, "model.duration");
-                  $contains(widgets.model, "turbo")
-                    ? (
-                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
-                    : (
-                        $rate := $lookup({"720p": 0.15, "1080p": 0.16}, $res);
-                        {"type":"usd","usd": $rate * $d}
-                      )
-                )
-                """,
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        model: dict,
-        first_frame: Input.Image,
-        end_frame: Input.Image,
-        prompt: str,
-        seed: int,
-    ) -> IO.NodeOutput:
-        validate_string(prompt, max_length=2000)
-        validate_images_aspect_ratio_closeness(first_frame, end_frame, min_rel=0.8, max_rel=1.25, strict=False)
-        payload = TaskCreationRequest(
-            model=model["model"],
-            prompt=prompt,
-            duration=model["duration"],
-            seed=seed,
-            resolution=model["resolution"],
-            audio=model["audio"],
-            images=[
-                (await upload_images_to_comfyapi(cls, frame, max_images=1, mime_type="image/png"))[0]
-                for frame in (first_frame, end_frame)
-            ],
-        )
-        results = await execute_task(cls, VIDU_START_END_VIDEO, payload)
-        return IO.NodeOutput(await download_url_to_video_output(results[0].url))
-
-
 class ViduExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -1720,7 +1497,6 @@ class ViduExtension(ComfyExtension):
             ViduMultiFrameVideoNode,
             Vidu3TextToVideoNode,
             Vidu3ImageToVideoNode,
-            Vidu3StartEndToVideoNode,
         ]
 
 

comfy_api_nodes/nodes_wan.py

@@ -227,14 +227,12 @@ class WanTextToImageApi(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -357,7 +355,6 @@ class WanImageToImageApi(IO.ComfyNode):
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -498,21 +495,18 @@ class WanTextToVideoApi(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="If no audio input is provided, generate audio automatically.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "prompt_extend",
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "shot_type",
@@ -521,7 +515,6 @@ class WanTextToVideoApi(IO.ComfyNode):
                     "single continuous shot or multiple shots with cuts. "
                     "This parameter takes effect only when prompt_extend is True.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -674,21 +667,18 @@ class WanImageToVideoApi(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="If no audio input is provided, generate audio automatically.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "prompt_extend",
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
-                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "shot_type",
@@ -697,7 +687,6 @@ class WanImageToVideoApi(IO.ComfyNode):
                     "single continuous shot or multiple shots with cuts. "
                     "This parameter takes effect only when prompt_extend is True.",
                     optional=True,
-                    advanced=True,
                 ),
             ],
             outputs=[
@@ -850,13 +839,11 @@ class WanReferenceVideoApi(IO.ComfyNode):
                     options=["single", "multi"],
                     tooltip="Specifies the shot type for the generated video, that is, whether the video is a "
                     "single continuous shot or multiple shots with cuts.",
-                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
-                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/util/__init__.py

@@ -9,8 +9,6 @@ from .client import (
 from .conversions import (
     audio_bytes_to_audio_input,
     audio_input_to_mp3,
-    audio_ndarray_to_bytesio,
-    audio_tensor_to_contiguous_ndarray,
     audio_to_base64_string,
     bytesio_to_image_tensor,
     convert_mask_to_image,
@@ -80,8 +78,6 @@ __all__ = [
     # Conversions
     "audio_bytes_to_audio_input",
     "audio_input_to_mp3",
-    "audio_ndarray_to_bytesio",
-    "audio_tensor_to_contiguous_ndarray",
     "audio_to_base64_string",
     "bytesio_to_image_tensor",
     "convert_mask_to_image",

comfy_extras/nodes_advanced_samplers.py

@@ -47,8 +47,8 @@ class SamplerLCMUpscale(io.ComfyNode):
             node_id="SamplerLCMUpscale",
             category="sampling/custom_sampling/samplers",
             inputs=[
-                io.Float.Input("scale_ratio", default=1.0, min=0.1, max=20.0, step=0.01, advanced=True),
-                io.Int.Input("scale_steps", default=-1, min=-1, max=1000, step=1, advanced=True),
+                io.Float.Input("scale_ratio", default=1.0, min=0.1, max=20.0, step=0.01),
+                io.Int.Input("scale_steps", default=-1, min=-1, max=1000, step=1),
                 io.Combo.Input("upscale_method", options=cls.UPSCALE_METHODS),
             ],
             outputs=[io.Sampler.Output()],
@@ -94,7 +94,7 @@ class SamplerEulerCFGpp(io.ComfyNode):
             display_name="SamplerEulerCFG++",
             category="_for_testing",  # "sampling/custom_sampling/samplers"
             inputs=[
-                io.Combo.Input("version", options=["regular", "alternative"], advanced=True),
+                io.Combo.Input("version", options=["regular", "alternative"]),
             ],
             outputs=[io.Sampler.Output()],
             is_experimental=True,

comfy_extras/nodes_apg.py

@@ -26,7 +26,6 @@ class APG(io.ComfyNode):
                     max=10.0,
                     step=0.01,
                     tooltip="Controls the scale of the parallel guidance vector. Default CFG behavior at a setting of 1.",
-                    advanced=True,
                 ),
                 io.Float.Input(
                     "norm_threshold",
@@ -35,7 +34,6 @@ class APG(io.ComfyNode):
                     max=50.0,
                     step=0.1,
                     tooltip="Normalize guidance vector to this value, normalization disable at a setting of 0.",
-                    advanced=True,
                 ),
                 io.Float.Input(
                     "momentum",
@@ -44,7 +42,6 @@ class APG(io.ComfyNode):
                     max=1.0,
                     step=0.01,
                     tooltip="Controls a running average of guidance during diffusion, disabled at a setting of 0.",
-                    advanced=True,
                 ),
             ],
             outputs=[io.Model.Output()],

comfy_extras/nodes_attention_multiply.py

@@ -28,10 +28,10 @@ class UNetSelfAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,
@@ -51,10 +51,10 @@ class UNetCrossAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,
@@ -75,10 +75,10 @@ class CLIPAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Clip.Input("clip"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
             ],
             outputs=[io.Clip.Output()],
             is_experimental=True,
@@ -109,10 +109,10 @@ class UNetTemporalAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("self_temporal", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("cross_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
-                io.Float.Input("cross_temporal", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("self_temporal", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("cross_structural", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("cross_temporal", default=1.0, min=0.0, max=10.0, step=0.01),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,