Chain-of-thought reasoning for visual design.
Zen Designer Thinking is a vision-language model that applies chain-of-thought reasoning to design problems. Decomposes visual design decisions into explicit reasoning steps — analyzing layout, color theory, typography, and accessibility before producing recommendations.
| Property | Value |
|---|---|
| Parameters | 4B |
| Context | 32K |
| Modalities | Text + Vision |
| License | Apache 2.0 |
from transformers import AutoModelForCausalLM, AutoProcessor
model = AutoModelForCausalLM.from_pretrained("zenlm/zen-designer-thinking", trust_remote_code=True)
processor = AutoProcessor.from_pretrained("zenlm/zen-designer-thinking", trust_remote_code=True)
from PIL import Image
image = Image.open("landing-page.png")
messages = [{"role": "user", "content": [
{"type": "image"},
{"type": "text", "text": "Think step by step: evaluate the color contrast, visual hierarchy, and accessibility of this landing page design."}
]}]
inputs = processor(images=image, text=processor.apply_chat_template(messages), return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=1024)
print(processor.decode(output[0], skip_special_tokens=True))- zen-designer-instruct — Design instruction variant
- zen-eco-thinking — General chain-of-thought model
- Zen LM — Full model family