This paper shows how a video generator can improve its own videos during sampling, without extra training or outside checkers.
The paper turns video avatars from passive puppets into active doers that can plan, act, check their own work, and fix mistakes over many steps.