Motion Attribution for Video Generation

Despite the rapid progress of video generation models, the role of data in influencing motion is poorly understood. We present Motive (MOTIon attribution for Video gEneration), a motion-centric, gradient-based data attribution framework that scales to modern, large, high-quality video datasets and models. We use this to study which fine-tuning clips improve or degrade temporal dynamics. Motive isolates temporal dynamics from static appearance via motion-weighted loss masks, yielding efficient and scalable motion-specific influence computation. On text-to-video models, Motive identifies clips that strongly affect motion and guides data curation that improves temporal consistency and physical plausibility. With Motive-selected high-influence data, we improve both motion smoothness and dynamic degree on VBench, achieving a 74.1% human preference win rate compared with the pretrained base model. To our knowledge, this is the first framework to attribute motion rather than visual appearance in video generative models and to use it to curate fine-tuning data.

AI video generators are getting impressive, but their motion still looks off: characters drift, objects fly by impossibly, and the physics breaks down in ways that humans immediately notice. We have no clear way to ask which training videos taught a model these bad habits, or which ones could fix them. Most existing tools for tracing how training data shapes outputs were built for still images, where there is no motion to explain. We built MOTIVE, the first method that traces a generated video's motion back to the specific training clips that influenced it, independent of the scene's appearance. MOTIVE focuses on the moving parts of each video and asks how strongly each training clip pushed the model toward producing that kind of motion. Because the scores are motion-specific, we can pick out the small slice of training data, about one clip in a hundred, that most improves how the model moves. When we retrained a popular open-source video generator on just that top slice, human viewers preferred its motion 74% of the time. MOTIVE gives video-model developers a practical way to curate better training sets, producing smoother, more physically plausible videos without requiring additional data collection.