Research map / Video / Video generation
PhyCo: Learning Controllable Physical Priors for Generative Motion
Sriram Narayanan, Ziyu Jiang, Srinivasa Narasimhan, Manmohan Chandraker
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 |
|---|---|---|---|
| fine-tune on PhyCo data with text only | Cosmos-Predict2-2B · Physics-IQ (120 frames extrapolated) | IQ score: 27.7 -> 30.9 | Improved |
| add physical property maps via ControlNet (ablation) | PhyCo text-only · Physics-IQ (120 frames extrapolated) | IQ score: 30.9 -> 35.3 | Improved |
| add physical property maps via ControlNet (ablation) | PhyCo text-only · Physics-IQ (57 frames + last-frame repetition) | IQ score: 36.5 -> 38.9 | Improved |
| add VLM reward loss (ablation) | PhyCo ControlNet · Physics-IQ (120 frames extrapolated) | IQ score: 35.3 -> 36.3 | Improved |
| add VLM reward loss (ablation) | PhyCo ControlNet · Physics-IQ (57 frames + last-frame repetition) | IQ score: 38.9 -> 43.6 | Improved |
| add VLM reward loss (ablation) | PhyCo ControlNet · Physics-IQ (120 frames extrapolated) | Thermodynamics: 40.1 -> 33.1 | Hurt |
| add VLM reward loss (ablation) | PhyCo ControlNet · PhyCo synthetic test set (100 videos) | Force direction deviation (deg): 38.05 -> 22.53 | Improved |
| add VLM reward loss (ablation) | PhyCo ControlNet · PhyCo synthetic test set (100 videos) | Restitution error: 0.28 -> 0.16 | Improved |
| add property maps via ControlNet (ablation) | PhyCo text-only · PhyCo synthetic test set (100 videos) | Friction error: 0.30 -> 0.24 | Improved |
| add property maps via ControlNet (ablation) | PhyCo text-only · PhyCo synthetic test set (100 videos) | Force magnitude error: 0.31 -> 0.33 | Hurt |
| add VLM reward loss (ablation) | PhyCo ControlNet · Physics-IQ FVMD | Optics FVMD (lower better): 2991.0 -> 3032.9 | Hurt |
| add VLM reward loss (ablation) | PhyCo ControlNet · Physics-IQ FVMD | Solid mechanics FVMD (lower better): 2340.0 -> 2337.7 | No clear effect |
Closest papers by meaning
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.94)
- PhysVid: Physics Aware Local Conditioning for Generative Video Models (CVPR 2026 · similarity 0.92)
- PhyMAGIC: Physical Motion-Aware Generative Inference with Confidence-guided VLM (ECCV 2026 · similarity 0.92)
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · similarity 0.89)
- Hierarchical Fine-grained Preference Optimization for Physically Plausible Video Generation (NeurIPS 2025 · similarity 0.89)
- PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation (NeurIPS 2026 · similarity 0.89)
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.
