Research map / Image generation / Text-to-image diffusion
CoCo: Code as CoT for Text-to-Image Preview and Rare Concept Generation
Li Haodong, Chunmei Qing, Huanyu Zhang, Dongzhi Jiang, Yihang Zou, Hongbo Peng, Dingming Li, Yuhong Dai, ZePeng Lin, Juanxi Tian, Yi Zhou, Siqi Dai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| generate executable code, render draft, then refine (Code-as-CoT) instead of direct generation | Bagel · StructT2IBench | overall accuracy (%): 4.69 -> 73.52 | Improved |
| use code as CoT instead of text CoT | Bagel · StructT2IBench | overall accuracy (%): Bagel-Think 9.03 -> CoCo 73.52 | Improved |
| Code-as-CoT on text rendering | Bagel · OneIG-Bench | overall score: 0.305 -> 0.853 | Improved |
| Code-as-CoT on long text rendering | Bagel · LongText-Bench | overall score: 0.342 -> 0.754 | Improved |
| add natural-language thinking to Bagel | Bagel · LongText-Bench | overall score: 0.342 -> 0.087 | Hurt |
| add Text-Code supervision (fine-tune to generate executable code) (ablation) | CoCo · LongText-Bench | code compile success rate: 9.06% -> 100% | Improved |
| reduce proportion of Text-Code supervision in the training mixture (best r_c=0.05) (ablation) | CoCo · LongText-Bench | score EN/CN: 0.755 EN, 0.753 CN at r_c=0.05 | Improved |
| Code-as-CoT vs closed models on Puzzle and Science | CoCo · StructT2IBench Puzzle | accuracy (%): CoCo 49.10 vs Nano banana 63.87 | Hurt |
Closest papers by meaning
- Uni-CoT: Towards Unified Chain-of-Thought Reasoning Across Text and Vision (ICLR 2026 · similarity 0.90)
- T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT (NeurIPS 2025 · similarity 0.89)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (NeurIPS 2025 · similarity 0.89)
- CoT-lized Diffusion: Let's Reinforce T2I Generation Step-by-step (NeurIPS 2025 · similarity 0.89)
- GIR-Bench: Versatile Benchmark for Generating Images with Reasoning (ICLR 2026 · similarity 0.88)
- TVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding (ICML 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
