Research map / Video / Video generation
Physics-Guided Motion Loss for Video Generation Model
Bowen Xue, Giuseppe Claudio Guarnera, Shuang Zhao, Zahra Montazeri
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 frequency-domain physics motion loss (translation, rotation, scaling) | Open-Sora · OpenVID-1M | Action Recognition Score: 60.77 -> 69.71 | Improved |
| add frequency-domain physics motion loss | Open-Sora · OpenVID-1M | Warping Error: 0.0089 -> 0.0056 | Improved |
| add frequency-domain physics motion loss | MVDIT · OpenVID-1M | Motion Accuracy Score: 44.0 -> 51.0 | Improved |
| add frequency-domain physics motion loss with LoRA | Hunyuan · OpenVID-1M | Action Recognition Score: 68.93 -> 73.15 | Improved |
| add standard LoRA fine-tune with no extra loss (ablation) | Hunyuan · OpenVID-1M | Action Recognition Score: 68.93 -> 69.02 | No clear effect |
| use optical-flow consistency loss (RAFT) instead of spectral loss (ablation) | Hunyuan LoRA · OpenVID-1M | Motion Accuracy Score: 56.0 -> 56.0 | No clear effect |
| spectral loss instead of flow loss (ablation) | Hunyuan LoRA · OpenVID-1M | Warping Error: 0.0022 -> 0.0016 | Improved |
| add translation loss term (ablation) | Open-Sora physics loss · OpenVID-1M | Warping Error: 0.0078 -> 0.0056 | Improved |
| add rotation loss term (ablation) | Open-Sora physics loss · OpenVID-1M | Action Recognition Score: 66.50 -> 69.71 | Improved |
| add scaling loss term (ablation) | Open-Sora physics loss · OpenVID-1M | Motion Accuracy Score: 44.00 -> 49.00 | Improved |
| apply physics loss on complex-motion prompts | Open-Sora/Hunyuan-style baseline · OpenVID-1M complex subset | Warping Error: 0.0020 -> 0.0011 | Improved |
| apply physics loss on simple-motion prompts | baseline · OpenVID-1M simple subset | Warping Error: 0.0025 -> 0.0018 | Improved |
Closest papers by meaning
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · similarity 0.90)
- Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models (AAAI 2025 · similarity 0.89)
- Video Diffusion Models are Training-free Motion Interpreter and Controller (NeurIPS 2024 · similarity 0.88)
- MotionCraft: Physics-Based Zero-Shot Video Generation (NeurIPS 2024 · similarity 0.88)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.88)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (NeurIPS 2026 · 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.
