We propose FreeLong, a straightforward and training-free approach to extend an existing short video diffusion model for consistent long video generation. In this repository, we utilize LaVie as a case ...