Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection

Feng Liu (University of Melbourne) · Bo Han (HKBU / RIKEN) · ZiHao Lian (South China University of Technology) · Mingkui Tan (South China University of Technology) · Shuhai Zhang (South China University of Technology) · Jiahao Yang (South China University of Technology) · Daiyuan Li (South China University of Technology) · Guoxuan Pang (University of Science and Technology of China) · Shutao Li (Hunan University)

ai-generated videosdetection mechanismsdiffusion models distributional shiftsfeature distancesmaximum mean discrepancymotion-aware temporal modelingnormalized spatiotemporal gradientnsg estimatornsg-based video detectionperformance validationprobability flow conservationspatial probability gradientsspatiotemporal dynamicstemporal density changesvisual realism

AI-generated videos have achieved near-perfect visual realism (e.g., Sora), urgently necessitating reliable detection mechanisms. However, detecting such videos faces significant challenges in modeling high-dimensional spatiotemporal dynamics and identifying subtle anomalies that violate physical laws. In this paper, we propose a physics-driven AI-generated video detection paradigm based on probability flow conservation principles. Specifically, we propose a statistic called Normalized Spatiotemporal Gradient (NSG), which quantifies the ratio of spatial probability gradients to temporal density changes, explicitly capturing deviations from natural video dynamics. Leveraging pre-trained diffusion models, we develop an NSG estimator through spatial gradients approximation and motion-aware temporal modeling without complex motion decomposition while preserving physical constraints. Building on this, we propose an NSG-based video detection method (NSG-VD) that computes the Maximum Mean Discrepancy (MMD) between NSG features of the test and real videos as a detection metric. Last, we derive an upper bound of NSG feature distances between real and generated videos, proving that generated videos exhibit amplified discrepancies due to distributional shifts. Extensive experiments confirm that NSG-VD outperforms state-of-the-art baselines by 16.00\% in Recall and 10.75\% in F1-Score, validating the superior performance of NSG-VD. The source code is available at \url{https://github.com/ZSHsh98/NSG-VD}.