Research map / Image generation / Text-to-image diffusion
Powerful and Flexible: Personalized Text-to-Image Generation via Reinforcement Learning
Fanyue Wei, Wei Zeng, Zhenyang Li, Dawei Yin, Lixin Duan, Wen Li
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 Look Forward reward via DPG framework to DreamBooth | DreamBooth · DreamBooth benchmark | DINO: 0.694 -> 0.738 | Improved |
| add Look Forward reward to DreamBooth | DreamBooth · DreamBooth benchmark | CLIP-T: 0.282 -> 0.269 | Hurt |
| add DINO reward to DreamBooth (ablation) | DreamBooth · DreamBooth benchmark | DINO: 0.694 -> 0.723 | Improved |
| add Look Forward reward to Custom Diffusion | Custom Diffusion · DreamBooth benchmark | DINO: 0.649 -> 0.669 | Improved |
| add DINO reward to Custom Diffusion (ablation) | Custom Diffusion · DreamBooth benchmark | DINO: 0.649 -> 0.640 | No clear effect |
| add Look Forward reward to DreamBooth | DreamBooth · Custom benchmark | DINO: 0.640 -> 0.680 | Improved |
| Look Forward vs DINO reward on Custom benchmark (ablation) | DreamBooth · Custom benchmark | DINO: 0.653 -> 0.680 | Improved |
| increase DINO reward weight from 0.1 to 1 (ablation) | DreamBooth + DPG · robot_toy concept | DINO / CLIP-T: DINO 0.644 -> 0.727 ; CLIP-T 0.239 -> 0.211 | Mixed |
| remove discount rate gamma (no discount) (ablation) | DreamBooth + DPG · clock concept | DINO: 0.704 -> 0.727 | Improved |
| DPG approach vs human preference on DreamBooth | DreamBooth · User study | text fidelity preference: 19.6% vs 20.4%, similar 60.0% | No clear effect |
Closest papers by meaning
- PaRa: Personalizing Text-to-Image Diffusion via Parameter Rank Reduction (ICLR 2025 · similarity 0.90)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.90)
- Dynamic Prompt Optimizing for Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- JeDi: Joint-Image Diffusion Models for Finetuning-Free Personalized Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.89)
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 2025 · 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.
