PhysVid: Physics Aware Local Conditioning for Generative Video Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912987497889792 |
|---|---|
| author | Pathak, Saurabh Arani, Elahe Pechenizkiy, Mykola Zonooz, Bahram |
| author_facet | Pathak, Saurabh Arani, Elahe Pechenizkiy, Mykola Zonooz, Bahram |
| contents | Generative video models achieve high visual fidelity but often violate basic physical principles, limiting reliability in real-world settings. Prior attempts to inject physics rely on conditioning: frame-level signals are domain-specific and short-horizon, while global text prompts are coarse and noisy, missing fine-grained dynamics. We present PhysVid, a physics-aware local conditioning scheme that operates over temporally contiguous chunks of frames. Each chunk is annotated with physics-grounded descriptions of states, interactions, and constraints, which are fused with the global prompt via chunk-aware cross-attention during training. At inference, we introduce negative physics prompts (descriptions of locally relevant law violations) to steer generation away from implausible trajectories. On VideoPhy, PhysVid improves physical commonsense scores by $\approx 33\%$ over baseline video generators, and by up to $\approx 8\%$ on VideoPhy2. These results show that local, physics-aware guidance substantially increases physical plausibility in generative video and marks a step toward physics-grounded video models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_26285 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | PhysVid: Physics Aware Local Conditioning for Generative Video Models Pathak, Saurabh Arani, Elahe Pechenizkiy, Mykola Zonooz, Bahram Computer Vision and Pattern Recognition Artificial Intelligence I.2.6; I.2.10; I.5.1 Generative video models achieve high visual fidelity but often violate basic physical principles, limiting reliability in real-world settings. Prior attempts to inject physics rely on conditioning: frame-level signals are domain-specific and short-horizon, while global text prompts are coarse and noisy, missing fine-grained dynamics. We present PhysVid, a physics-aware local conditioning scheme that operates over temporally contiguous chunks of frames. Each chunk is annotated with physics-grounded descriptions of states, interactions, and constraints, which are fused with the global prompt via chunk-aware cross-attention during training. At inference, we introduce negative physics prompts (descriptions of locally relevant law violations) to steer generation away from implausible trajectories. On VideoPhy, PhysVid improves physical commonsense scores by $\approx 33\%$ over baseline video generators, and by up to $\approx 8\%$ on VideoPhy2. These results show that local, physics-aware guidance substantially increases physical plausibility in generative video and marks a step toward physics-grounded video models. |
| title | PhysVid: Physics Aware Local Conditioning for Generative Video Models |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence I.2.6; I.2.10; I.5.1 |
| url | https://arxiv.org/abs/2603.26285 |