Research map / Image generation / Text-to-image diffusion
UniGen-1.5: Enhancing Image Generation and Editing through Reward Unification in RL
Rui Tian, Mingfei Gao, Haiming Gang, Jiasen Lu, Zhe Gan, Yinfei Yang, Zuxuan Wu, Afshin Dehghan
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 unified GRPO RL stage after Post-SFT (ablation) | UniGen-1.5 · ImgEdit | overall score: 3.93 -> 4.31 | Improved |
| add unified GRPO RL stage after Post-SFT (ablation) | UniGen-1.5 · GenEval | overall: 0.85 -> 0.89 | Improved |
| add unified GRPO RL stage after Post-SFT (ablation) | UniGen-1.5 · DPG-Bench | overall: 84.19 -> 86.83 | Improved |
| include image editing in RL alongside text-to-image (ablation) | UniGen-1.5 RL · ImgEdit | overall score: 4.01 -> 4.31 | Improved |
| include image editing in RL alongside text-to-image (ablation) | UniGen-1.5 RL · GenEval | overall: 0.90 -> 0.89 | No clear effect |
| include text-to-image in RL alongside image editing (ablation) | UniGen-1.5 RL · GenEval | overall: 0.85 -> 0.89 | Improved |
| include text-to-image in RL alongside image editing (ablation) | UniGen-1.5 RL · ImgEdit | overall score: 4.32 -> 4.31 | No clear effect |
| add Edit Instruction Alignment stage before RL (ablation) | UniGen-1.5 · ImgEdit (before RL) | overall score: 3.87 -> 3.93 | Improved |
| add Edit Instruction Alignment stage before RL (ablation) | UniGen-1.5 · ImgEdit (after RL) | overall score: 4.08 -> 4.31 | Improved |
| add Edit Instruction Alignment stage before RL (ablation) | UniGen-1.5 · GenEval (after RL) | overall: 0.90 -> 0.89 | No clear effect |
| add Edit Instruction Alignment stage before RL (ablation) | UniGen-1.5 · DPG-Bench (after RL) | overall: 86.96 -> 86.83 | No clear effect |
| RL gain on ImgEdit without Edit Instruction Alignment (ablation) | UniGen-1.5 · ImgEdit | overall score: 3.87 -> 4.08 | Improved |
Closest papers by meaning
- UniGen: Enhanced Training & Test-Time Strategies for Unified Multimodal Understanding and Generation (NeurIPS 2025 · similarity 0.90)
- RetouchIQ: MLLM Agents for Instruction-Based Image Retouching with Generalist Reward (CVPR 2026 · similarity 0.90)
- The Promise of RL for Autoregressive Image Editing (NeurIPS 2025 · similarity 0.89)
- Edit-R2: Context-Aware Reinforcement Learning for Multi-Turn Image Editing (NeurIPS 2026 · similarity 0.88)
- Janus-Pro-R1: Advancing Collaborative Visual Comprehension and Generation via Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Co-Reinforcement Learning for Unified Multimodal Understanding and Generation (NeurIPS 2025 · 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.
