Zen Designer Instruct

Vision-language model for design instruction following.

Overview

Zen Designer Instruct is a vision-language model specialized for design tasks. Understands visual layouts, UI components, typography, and color systems. Follows design instructions to generate specifications, critique compositions, and produce structured design output.

Property	Value
Parameters	4B
Context	32K
Modalities	Text + Vision
License	Apache 2.0

Usage

from transformers import AutoModelForCausalLM, AutoProcessor

model = AutoModelForCausalLM.from_pretrained("zenlm/zen-designer-instruct", trust_remote_code=True)
processor = AutoProcessor.from_pretrained("zenlm/zen-designer-instruct", trust_remote_code=True)

from PIL import Image
image = Image.open("mockup.png")

messages = [{"role": "user", "content": [
    {"type": "image"},
    {"type": "text", "text": "Analyze this UI design. Suggest improvements for visual hierarchy and spacing."}
]}]

inputs = processor(images=image, text=processor.apply_chat_template(messages), return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(output[0], skip_special_tokens=True))

Related

• zen-designer-thinking — Design reasoning variant
• zen-eco — Base 4B model
• Zen LM — Full model family

Apache 2.0 · Zen LM · Hanzo AI