Research map / Video / Video generation
PhysAlign: Learning Physical Priors for Dynamical Event-Driven Video Generation via Representation Alignment
Mengxian Li, Zhan Wang, Fan Qi, Changsheng Xu
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 Physical Dynamics Extractor (PDE) (ablation) | PhysAlign · EventPhy | PC: 29.05 -> 34.83 | Improved |
| add Event-Aware Temporal Gating (ETG) (ablation) | PhysAlign · EventPhy | PC: 32.23 -> 34.83 | Improved |
| add conventional REPA to the baseline (ablation) | Wan2.2-TI2V-5B · EventPhy | PC: 23.03 -> 28.87 | Improved |
| add basic LoRA fine-tuning to the baseline (ablation) | Wan2.2-TI2V-5B · EventPhy | PC: 23.03 -> 26.30 | Improved |
| use full PhysAlign over baseline+REPA (ablation) | Wan2.2-TI2V-5B · EventPhy | SA: 15.64 -> 16.35 | Improved |
| align at layer 18 instead of layer 10 (ablation) | PhysAlign · EventPhy | PC: 24.38 -> 34.83 | Improved |
| align at layer 18 instead of layer 26 (ablation) | PhysAlign · EventPhy | PC: 23.29 -> 34.83 | Improved |
| align at layer 22 instead of layer 18 (ablation) | PhysAlign · EventPhy | PC: 34.83 -> 25.09 | Hurt |
| PhysAlign vs VLIPP on VideoPhy | PhysAlign · VideoPhy | SA: 51.17 -> 53.98 | Improved |
| PhysAlign vs PhyT2V on VideoPhy2 | PhysAlign · VideoPhy2 | SA: 26.67 -> 23.67 | Hurt |
| PhysAlign vs VLIPP on VBench quality score | PhysAlign · VBench | Quality Score: 75.22 -> 79.05 | Improved |
| PhysAlign vs VideoREPA on aesthetic quality | PhysAlign · VBench | Aesthetic Quality: 50.40 -> 48.92 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VideoREPA-5B — VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 2025) · better on 7 of 8 shared results
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · 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
- PhysVid: Physics Aware Local Conditioning for Generative Video Models (CVPR 2026 · similarity 0.91)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (NeurIPS 2026 · similarity 0.91)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.91)
- Hierarchical Fine-grained Preference Optimization for Physically Plausible Video Generation (NeurIPS 2025 · similarity 0.89)
- VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 2025 · similarity 0.88)
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · 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.
