Research map / Image generation / Text-to-image diffusion
RealGen: Photorealistic Text-to-Image Generation via Detector-Guided Rewards
ye junyan, Leqi Zhu, Yuncheng Guo, Dongzhi Jiang, Zilong Huang, Yifan Zhang, Zhiyuan Yan, Haohuan Fu, Conghui He, Weijia 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 |
|---|---|---|---|
| use Detector-Score reward instead of no RL (baseline FLUX.1-dev) (ablation) | FLUX.1-dev · RealBench | Effort: 14.85 -> 31.71 | Improved |
| use Detector-Score reward instead of no RL (ablation) | FLUX.1-dev · RealBench | GPT 5-Prompt: 43.03 -> 92.79 | Improved |
| use Detector-Score reward instead of no RL (ablation) | FLUX.1-dev · RealBench | VS Real: 12.61 -> 43.41 | Improved |
| use Detector-Score reward instead of no RL (ablation) | FLUX.1-dev · RealBench | CLIP: 86.33 -> 86.80 | Improved |
| use PickScore reward instead of Detector-Score (ablation) | FLUX.1-dev · RealBench | GPT 5-Prompt: 92.79 -> 45.15 | Hurt |
| use PickScore reward instead of Detector-Score (ablation) | FLUX.1-dev · RealBench | Effort: 31.71 -> 12.75 | Hurt |
| use HPSv2.1 reward instead of Detector-Score (ablation) | FLUX.1-dev · RealBench | GPT 5-Prompt: 92.79 -> 39.00 | Hurt |
| use HPSv2.1 reward instead of Detector-Score (ablation) | FLUX.1-dev · RealBench | VS Real: 43.41 -> 19.12 | Hurt |
| add LLM prompt optimization (Phase 1) to RealGen (ablation) | RealGen · RealBench | VS Real: 43.41 -> 50.15 | Improved |
| add LLM prompt optimization to RealGen (ablation) | RealGen · RealBench | PickScore: 23.58 -> 21.75 | Hurt |
| add LLM prompt optimization to RealGen (ablation) | RealGen · RealBench | HPSv3: 13.61 -> 11.11 | Hurt |
| add only Phase 1 LLM prompt optimization before diffusion optimization (ablation) | RealGen · RealBench | GPT-prompt score, VS Real | Improved |
Closest papers by meaning
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.88)
- RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.86)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.86)
- RealCustom: Narrowing Real Text Word for Real-Time Open-Domain Text-to-Image Customization (CVPR 2024 · similarity 0.86)
- PixArt-$\alpha$: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis (ICLR 2024 · similarity 0.86)
- RealRAG: Retrieval-augmented Realistic Image Generation via Self-reflective Contrastive Learning (ICML 2025 · similarity 0.86)
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.
