Research map / Image generation / Text-to-image diffusion
Think-Then-Generate: Reasoning-Aware Text-to-Image Diffusion with LLM Encoders
Siqi Kou, Jiachun Jin, Zetong Zhou, Ye Ma, Yugang Wang, Quan Chen, Peng Jiang, Xiao Yang, Jun Zhu, Kai Yu, Zhijie Deng
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 |
|---|---|---|---|
| add zero-shot CoT system prompt to Qwen-Image encoder (ablation) | Qwen-Image · WISE | overall score: 0.61 -> 0.65 | Improved |
| add SFT on Gemini-2.5 CoT/rewrite data (ablation) | Qwen-Image · WISE | overall score: 0.65 -> 0.74 | Improved |
| add Dual-GRPO after SFT (ablation) | Qwen-Image · WISE | overall score: 0.74 -> 0.79 | Improved |
| add Dual-GRPO after SFT (ablation) | Qwen-Image · T2I-ReasonBench | overall accuracy: 66.1 -> 68.3 | Improved |
| add Dual-GRPO after SFT (ablation) | Qwen-Image-Edit · UniREditBench | score: 61.1 -> 68.7 | Improved |
| add Dual-GRPO after SFT (ablation) | Qwen-Image-Edit · RISEBench | score: 20.2 -> 23.9 | Improved |
| apply GRPO without SFT (ablation) | Qwen-Image · WISE | overall score: 0.65 -> 0.70 | Improved |
| use Qwen3-8B instead of Gemini-2.5 to build SFT data (ablation) | Qwen-Image · WISE | overall score: 0.74 -> 0.71 | Hurt |
| use Qwen3-8B SFT data followed by Dual-GRPO (ablation) | Qwen-Image · WISE | overall score: 0.71 -> 0.76 | Improved |
| apply SFT on the LLM encoder (ablation) | Qwen2.5-VL encoder · latent space (t-SNE) | distribution overlap | No clear effect |
| apply full T2G to Qwen-Image | Qwen-Image · WISE | overall score: 0.61 -> 0.79 | Improved |
Closest papers by meaning
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.92)
- A Comprehensive Study of Decoder-Only LLMs for Text-to-Image Generation (CVPR 2025 · similarity 0.90)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.90)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.90)
- DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models (ICML 2026 · similarity 0.90)
- LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation (AAAI 2025 · similarity 0.90)
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.
