kapynResearch

VideoFlexTok: Flexible-Length Coarse-to-Fine Video Tokenization

VideoFlexTok tokenizes video flexibly for better downstream modeling. This novel approach moves beyond fixed 3D grids, allowing models to focus on essential information and reduce computational burden, especially for complex videos. It promises improved efficiency and performance in video generation tasks.

Apple ML Research·Jul 2, 2026

Opening Kapyn…