Research map / Image generation / Text-to-image diffusion
PromptEnhancer: Taming Your Rewriter for Text-to-Image Generation via Fine-Grained Reward
Linqing Wang, Zhiyong Xu, Ximing Xing, Yiji Cheng, Zhiyuan Zhao, Donghao Li, Tiankai Hang, Zhenxi Li, Jiale Tao, Qixun Wang, Ruihuang Li, Comi Chen and 6 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 PromptEnhancer rewriting | HY-Image 2.1 · GenEval | overall score: 0.80 -> 0.82 | Improved |
| add PromptEnhancer rewriting | Qwen-Image · GenEval | overall score: 0.84 -> 0.86 | Improved |
| add PromptEnhancer rewriting | Qwen-Image · T2I-CompBench | Spatial: 0.3222 -> 0.4472 | Improved |
| add PromptEnhancer rewriting | Qwen-Image · T2I-CompBench | 3D Spatial: 0.4146 -> 0.3943 | Hurt |
| add PromptEnhancer rewriting | HY-Image 2.1 · T2I-CompBench | Color/Shape/Texture: Color 0.7415 -> 0.8287; Shape 0.5889 -> 0.5884; Texture 0.7125 -> 0.7083 | Mixed |
| add BeautifulPrompt rewriting instead of PE | Qwen-Image · T2I-CompBench | Color: 0.7962 -> 0.5899 | Hurt |
| add BeautifulPrompt rewriting | HY-Image 2.1 · GenEval | overall score: 0.80 -> 0.81 | Improved |
| add SFT without CoT (ablation) | rewriter · T2I-Keypoints-Align | keypoint score: 81.0% -> 86.0% | Improved |
| add CoT supervision to SFT (ablation) | rewriter SFT · T2I-Keypoints-Align | keypoint score: 86.0% -> 85.29% | No clear effect |
| add GRPO with AlignEvaluator reward after SFT (ablation) | rewriter · T2I-Keypoints-Align | keypoint score: 85.29% -> 88.15% | Improved |
| add PromptEnhancer rewriting | base T2I model · T2I-Keypoints-Align (24 categories) | category accuracy: +5.1 points average; gains in 20 categories | Improved |
| add PromptEnhancer rewriting | base T2I model · T2I-Keypoints-Align | Text Layout / Interaction w/o Contact: -0.7 / -0.9 | Hurt |
Closest papers by meaning
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.91)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- Tailored Visions: Enhancing Text-to-Image Generation with Personalized Prompt Rewriting (CVPR 2024 · similarity 0.89)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.89)
- DreamDistribution: Learning Prompt Distribution for Diverse In-distribution Generation (ICLR 2025 · similarity 0.89)
- Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist Rewards (NeurIPS 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.
