🎩 DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generation

Official repository for the paper "DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generation".

[📖 Paper] [🤗 Model]

<p align="center"> <img src="figs/vis.jpg" width="100%"> <br> </p>

💥 News

• [2025.12.05] We release the arxiv paper. Code is coming soon. 🔥

🪄 Draft Before Generation

<p align="center"> <img src="figs/method.png" width="100%"> <br> </p>

We propose Draft-as-CoT (DraCo), a novel interleaved reasoning paradigm that fully leverages both textual and visual contents in CoT for better planning and verification.

Our method 🎨 first generates a low-resolution draft image as a preview, providing more concrete and structural visual planning and guidance.

Then, we 🔎 employ the model’s inherent understanding capability to verify potential semantic misalignments between the draft and input prompt, and 🖼️ perform refinement through selective corrections with superresolution.

🧠 Related Work

Explore our additional research on Autoregressive Text-to-Image Generation and CoT Reasoning

• [T2I-R1] T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT
• [ULMEvalKit] ULMEvalKit: One-Stop Eval ToolKit for Image Generation
• [Echo-4o] Echo-4o: Harnessing the Power of GPT-4o Synthetic Images for Improved Image Generation
• [Image Generation CoT] Can We Generate Images with CoT? Let's Verify and Reinforce Image Generation Step by Step?
• [Awesome-Nano-Banana-images] An Image Gallery Collecting Prompts to Create Stunning Images with Nano-banana