ICML oral Motive: first motion attribution framework for video generation, 74.1% win rate
cindy_x_wu · x · 2026-09-04
The ICML 2026 oral 'Motion Attribution for Video Generation' (Xindi Wu, Antonio Torralba, Sanja Fidler, Jonathan Lorraine, et al.) presents Motive, the first gradient-based framework attributing motion (rather than visual appearance) in video generation models. Using motion-weighted loss masks to isolate temporal dynamics, Motive identifies which fine-tuning clips improve or degrade motion and guides data curation, improving motion smoothness and dynamic degree on VBench with a 74.1% human preference win rate over the base model.
Related event: ICML oral paper on motion attribution for video generation released(2 posts)→
More from Multimodal
- This stunning W51A star-forming region image was AI-processed from raw JWST captures — fofrAI · 2026-10-21
- Purify Before You Align: filtering locally predictable info improves diffusion alignment — volokuleshov · 2026-10-21
- Mistral AI launches Voxtral 3, a real-time voice agent stack in 50+ languages — The Decoder · 2026-10-21
- Full AI Virtual Influencer Character Sheet for 'JataJett': Turnarounds, Expressions and Action Modes in One Page — CurieuxExplorer · 2026-10-21
- AI satire video: 'Paris City Tours, 50% off — slightly different experience' — DavidmComfort · 2026-10-21
- AI-Generated Series 'The Sage' Hits Episode 8, Made Entirely with Invideo Agent — Kavanthekid · 2026-10-21