PhysVid: Physics Aware Local Conditioning for Generative Video Models
This addresses the reliability issue in generative video models for real-world applications by enhancing physical plausibility, representing an incremental advance in physics-grounded video generation.
The paper tackled the problem of generative video models violating physical principles by introducing PhysVid, a physics-aware local conditioning scheme that uses temporally contiguous chunks with physics-grounded descriptions and negative prompts, resulting in improvements of approximately 33% on VideoPhy and up to 8% on VideoPhy2 in physical commonsense scores.
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.