Research map / Image generation / Text-to-image diffusion
RePrompt: Reasoning-Augmented Reprompting for Text-to-Image Generation via Reinforcement Learning
Mingrui Wu, Lu Wang, Pu Zhao, Fangkai Yang, Jianjin Zhang, Jianfeng Liu, Yuefeng Zhan, Weihao Han, Hao Sun, Jiayi Ji, Xiaoshuai Sun, Qingwei Lin and 4 more
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 reasoning trace to RL reprompting (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.68 -> 0.72 | Improved |
| add RL on top of SFT (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.69 -> 0.76 | Improved |
| add SFT on top of RL (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.72 -> 0.76 | Improved |
| use SFT only versus Qwen2.5 3B prompt (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.68 -> 0.69 | No clear effect |
| use ensemble ImageReward + VLM reward instead of ImageReward only (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.70 -> 0.72 | Improved |
| use ensemble reward instead of VLM reward only (ablation) | RePrompt · GenEval overall, FLUX | overall accuracy: 0.69 -> 0.72 | Improved |
| apply RePrompt to FLUX prompts | FLUX · GenEval position | position accuracy: 0.18 -> 0.62 | Improved |
| apply RePrompt to SD3 prompts (trained with FLUX) | SD3 · T2I-CompBench color | color score: 0.7941 -> 0.7866 | Hurt |
| apply RePrompt to FLUX | FLUX · GenEval single object | accuracy: 0.99 -> 0.98 | No clear effect |
| apply RePrompt to FLUX | FLUX · T2I-CompBench spatial | spatial score: 0.2494 -> 0.3301 | Improved |
| RePrompt vs Idea2Img optimization | FLUX · GenEval subset latency | latency: 140 -> 30 s/img | Improved |
Later papers that compared with this method
2 later accepted papers report RePrompt as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- PromptRL: Prompt Matters in RL for Flow-Based Image Generation (ICML 2026 · better on 1 of 1 shared result)
- POET: Preference Optimization for Enhanced Text-to-Image Generation (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GoT-R1: Unleashing Reasoning Capability of Autoregressive Visual Generation with Reinforcement Learning (ICLR 2026 · similarity 0.90)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.89)
- Preference Adaptive and Sequential Text-to-Image Generation (ICML 2025 · similarity 0.89)
- T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT (NeurIPS 2025 · similarity 0.89)
- Tailored Visions: Enhancing Text-to-Image Generation with Personalized Prompt Rewriting (CVPR 2024 · similarity 0.89)
- Benchmarking and Evolving Reason-Reflect-Rectify for Reflective Visual Generation (ICML 2026 · similarity 0.89)
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.
