Official repository for the paper "CoCo: CoCo as CoT for Text-to-Image Preview and Rare Concept Generation".
- [2026.3.09] We release the arxiv paper. Code is coming soon. 🔥
We propose CoCo-as-CoT (CoCo), a novel interleaved reasoning paradigm that fully leverages both textual and visual contents in CoT for better planning and verification.
Our method 🎨 first generates Python MatPlotlib Code and render by sandbox 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 🖼️ achieve higher fidelity and stronger semantic alignment.
Explore our additional research on Autoregressive Text-to-Image Generation and CoT Reasoning
- [GGEBench] GEBench: Benchmarking Image Generation Models as GUI Environments
- [VIBE] VIBE: A Systematic Benchmark for Visual Instruction-Driven Image Editing
- [GENIUS] GENIUS: Generative Fluid Intelligence Evaluation Suite
- [DraCo] DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generation
- [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