Research map / Video / Video generation
PhyGDPO: Physics-Aware Groupwise Direct Preference Optimization for Physically Consistent Text-to-Video Generation
Yuanhao Cai, Kunpeng Li, Menglin Jia, Jialiang Wang, Junzhe Sun, Feng Liang, Weifeng Chen, Felix Juefei-Xu, Chu Wang, Ali Thabet, Xiaoliang Dai, Xuan JU and 2 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 |
|---|---|---|---|
| post-train with PhyGDPO | Wan2.1-14B · VideoPhy2 | overall: 0.1288 -> 0.1627 | Improved |
| post-train with PhyGDPO | Wan2.1-14B · VideoPhy2 Hard | score: 0.0111 -> 0.0500 | Improved |
| add LoRA-SR (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1288 -> 0.1458 | Improved |
| add groupwise Plackett-Luce model (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1458 -> 0.1559 | Improved |
| add physics-guided rewarding (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1559 -> 0.1627 | Improved |
| add CoT data filtering in PhyAugPipe (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1475 -> 0.1525 | Improved |
| add action clustering in PhyAugPipe (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1525 -> 0.1575 | Improved |
| add physics-rewarded sampling in PhyAugPipe (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1575 -> 0.1627 | Improved |
| PhyGDPO instead of VideoDPO | Wan2.1-1.3B · VideoPhy2 | overall: 0.1305 -> 0.1407 | Improved |
| use LoRA-SR instead of full-copy reference (ablation) | PhyGDPO · VideoPhy2 | GPU memory: 48.7GB -> 25.3GB | Improved |
| use LoRA-SR instead of full-copy reference (ablation) | PhyGDPO · VideoPhy2 | overall: 0.1373 -> 0.1407 | Improved |
| DPO with LoRA-SR instead of LoRA-SFT (ablation) | Wan2.1-1.3B · VideoPhy2 | overall: 0.1283 -> 0.1407 | Improved |
Later papers that compared with this method
1 later accepted paper reports PhyGDPO as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- PhysPO: Physics-Aware Local Preference Optimization for Physically Consistent Video Diffusion (ECCV 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
- Hierarchical Fine-grained Preference Optimization for Physically Plausible Video Generation (NeurIPS 2025 · similarity 0.92)
- PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation (CVPR 2025 · similarity 0.89)
- VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 2025 · similarity 0.88)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.88)
- WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025 · similarity 0.88)
- Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation (ICML 2025 · similarity 0.87)
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.
