AnyFlow: Any-Step Video Diffusion Model with On-Policy Flow Map Distillation Paper โข 2605.13724 โข Published May 13 โข 104
HumanNet: Scaling Human-centric Video Learning to One Million Hours Paper โข 2605.06747 โข Published May 7 โข 53
MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head Paper โข 2601.07832 โข Published Jan 12 โข 53