AI generated videos detected by tracking motion consistency patterns
TRACE: Trajectory Representation and Consistency Estimation for AI-Generated Video Detection
Computer Vision and Pattern Recognition
Summary
Detecting videos made by AI is hard because fake videos often look very real and use different tricks. The authors found that real and AI videos behave differently when analyzed through a special tool that looks at their motion patterns, not just how they look. They created a method called TRACE that studies how movements change between video frames, which helps tell real videos apart from AI-generated ones, even if the AI used is new or unseen before. This makes the detection more reliable across many types of synthetic videos.
What this means in practice
- •For video security teams: Identify AI-generated videos across multiple synthesis models to prevent misinformation spread on video platforms.
- •For media verification services: Improve verification tools to flag synthetic videos regardless of their generation method for fact-checking workflows.$Commercial implications: Enables development of commercial video verification products that detect diverse AI-generated content robustly.
Authors
Huangsen Cao, Hongkang chu, Siyao Yu, Xin Ding, Jianfeng Dong, Yongwei Wang
Abstract
Recent advances in generative video models have enabled the synthesis of visually realistic content, posing significant challenges to synthetic video detection. Existing detectors often rely on appearance artifacts, semantic inconsistencies, and temporal patterns that may be generator-specific, limitating generalization to unseen synthesis models. We investigate whether responses to a pretrained generative model provide more transferable forensic cues. Our key observation is that real and AI-generated videos exhibit distinct \emph{velocity responses} under a pretrained Flow Matching video model. This distinction persists when different pretrained video-generation backbones are used as probes, suggesting that velocity responses offer transferable forensic signals beyond visual artificts. Motivated by this observation, we propose \textbf{TRACE} (\emph{\underline{T}rajectory \underline{R}epresentation \underline{a}nd \underline{C}onsistency \underline{E}stimation}), a generation-process-aware framework for AI-generated video detection. TRACE leverages a pretrained video DiT as a velocity-field probe to extract representations at multiple flow time points, and models cross-frame consistency through velocity differences between adjacent frames. We further introduce a \emph{Real-Centered Trajectory Optimization} objective that encourages generator-invariant representation learning. Extensive experiments on AIGVDBench demonstrate that TRACE generalizes effectively across diverse generators, substantially outperforming prior state-of-the-art methods on unseen open- and closed-source video generation models.