Research map / Image generation / Text-to-image diffusion
PromptRL: Prompt Matters in RL for Flow-Based Image Generation
Fu-Yun Wang, Han Zhang, Michaël Gharbi, Hongsheng Li, Taesung Park
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 jointly trained LM prompt refiner to flow RL | FlowGRPO · GenEval | overall score: 0.92 -> 0.97 | Improved |
| add inference-time prompt enhancement to PromptRL | PromptRL · GenEval | overall score: 0.94 -> 0.97 | Improved |
| apply naive prompt enhancement at inference to RL-tuned flow model | FlowGRPO · GenEval | overall score: 0.92 -> 0.81 | Hurt |
| apply naive prompt enhancement at inference to RL-tuned flow model | DiffusionNFT · GenEval | overall score: 0.88 -> 0.77 | Hurt |
| apply naive prompt enhancement (no joint training) to editing model | FLUX.1-Kontext · OmniEdit val | EditReward average: 1.19 -> 1.01 | Hurt |
| add PromptRL joint training to editing model | FLUX.1-Kontext · OmniEdit val | EditReward average: 1.19 -> 1.43 | Improved |
| train with 2 retained original prompts instead of 0 (ablation) | PromptRL · GenEval (no PE) | overall score: 0.83 -> 0.94 | Improved |
| train with 2 retained original prompts instead of 1 (ablation) | PromptRL · GenEval (no PE) | overall score: 0.77 -> 0.94 | Improved |
| retain 4 instead of 2 original prompts (ablation) | PromptRL · GenEval (no PE) | overall score: 0.94 -> 0.92 | Hurt |
| use single-reward instead of tag-based multi-reward training (ablation) | PromptRL · GenEval/OCR/PickScore | GenEval: 0.93 -> 0.97 | Mixed |
| scale flow-only RL to 2x rollouts instead of using PromptRL (ablation) | Flow-only RL · GenEval | overall score: 0.97 -> 0.93 | Hurt |
| apply PromptRL PE to unseen SD3 | SD3 · GenEval | overall score: 0.58 -> 0.77 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RePrompt — RePrompt: Reasoning-Augmented Reprompting for Text-to-Image Generation via Reinforcement Learning (ICLR 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
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 2025 · similarity 0.90)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.89)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.89)
- Not All Prompts Are Made Equal: Prompt-based Pruning of Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.88)
- The Promise of RL for Autoregressive Image Editing (NeurIPS 2025 · similarity 0.88)
- Dynamic Prompt Optimizing for Text-to-Image Generation (CVPR 2024 · 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.
