arXiv:2601.08828v2 Announce Type: replace-cross Abstract: 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 compu
The rapid progress in video generation models necessitates better understanding and control over their underlying mechanisms, particularly as they approach higher fidelity and broader application.
Improved interpretability and control over video generation models, through tools like Motive, are crucial for advancing AI capabilities and ensuring reliable, controllable outputs.
The ability to attribute generated motion to specific training data allows for targeted model improvement and debugging, moving beyond black-box optimization in video AI.
- · AI researchers
- · Video generation model developers
- · Creative industries using AI
- · AI ethics and safety researchers
- · Developers reliant on ad-hoc model tuning
- · Uninterpretable AI systems
More robust and controllable video generation models become available faster.
Improved model interpretability leads to accelerated development of realistic and application-specific video AI.
The development of highly reliable video AI could significantly disrupt media production, simulation, and advertising.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI