The Normalized Spatiotemporal Gradient (NSG) was jointly proposed in October 2025 by a research team from South China University of Technology, University of Science and Technology of China, and Pazhou Laboratory, among other universities and institutions. The relevant research results were published in a paper. Physics-Driven Spatiotemporal Modeling for AI-Generated Video DetectionIt was selected for NeurIPS 2025.

The NSG statistic quantifies the ratio of spatial probability gradient to temporal density change, explicitly capturing deviations from natural video dynamics. NSG captures the fundamental differences in how videos adhere to physical constraints while eliminating reliance on specific artifacts, enabling sensitive detection even when visual differences are imperceptible to humans or traditional models. Using a pre-trained diffusion model, researchers developed an NSG estimator through spatial gradient approximation and temporal modeling of motion awareness, preserving physical constraints without requiring complex motion decomposition.