🔥 FAR leverages clean visual context without additional image-to-video fine-tuning: Unconditional pretraining on UCF-101 achieves state-of-the-art results in both video generation (context frame = 0) ...

Customized video generation aims to produce videos featuring specific subjects under flexible user-defined conditions, yet existing methods often struggle with identity consistency and limited input ...