Research map / Image generation / Text-to-image diffusion
Scaling Multi-Reference Image Generation with Dynamic Reward Optimization
Wenwang Huang, Yusen Fu, Mengfei Huang, Junjie Wang, Yulin Li, Gan Liu, Jing Cai, Yancheng He, Zhuotao Tian
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 CSD style reward (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench objective style | style score: 0.36 -> 0.46 | Improved |
| add CSD style reward (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench MLLM style | style score: 6.33 -> 8.67 | Improved |
| add DAR (difficulty-aware advantage reweighting) over standard GRPO (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench objective average | average score: 0.69 -> 0.71 | Improved |
| add DAR over standard GRPO (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench MLLM average | average score: 8.03 -> 8.38 | Improved |
| add DRS (discriminative reward scaling) (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench MLLM style | style score: 8.12 -> 8.67 | Improved |
| add DRS (discriminative reward scaling) (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench objective average | average score: 0.71 -> 0.71 | No clear effect |
| add DRS (ablation) | DyRef on Qwen-Image-Edit-2511 · OmniRef-Bench MLLM aesthetics | Aes score: 8.03 -> 8.14 | Improved |
| apply DyRef to Qwen-Image-Edit-2511 | Qwen-Image-Edit-2511 · OmniRef-Bench MLLM average | average score: 4.97 -> 8.38 | Improved |
| apply DyRef to FLUX.2 [klein] 9B | FLUX.2 [klein] · OmniRef-Bench MLLM average | average score: 6.49 -> 8.03 | Improved |
| apply DyRef to Qwen-Image-Edit-2511 | Qwen-Image-Edit-2511 · ImgEdit | overall score: 4.07 -> 4.32 | Improved |
| apply DyRef to FLUX.2 [klein] | FLUX.2 [klein] · ImgEdit | overall score: 4.07 -> 4.11 | No clear effect |
| apply DyRef to Qwen-Image-Edit-2511 | Qwen-Image-Edit-2511 · DreamBench++ | CP*PF: 0.53 -> 0.62 | 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.
- DreamOmni2 — DreamOmni2: Multimodal Instruction-based Generation and Editing (CVPR 2026) · better on 2 of 2 shared results
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
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.85)
- Ctrl-U: Robust Conditional Image Generation via Uncertainty-aware Reward Modeling (ICLR 2025 · similarity 0.84)
- MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation? (NeurIPS 2025 · similarity 0.84)
- Improved Distribution Matching Distillation for Fast Image Synthesis (NeurIPS 2024 · similarity 0.84)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.84)
- Boost Your Human Image Generation Model via Direct Preference Optimization (CVPR 2025 · similarity 0.84)
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.
