VIDEO GENERATION WITH EFFICIENT COMPRESSION

Summary from the TLDR AI newsletter

CV-VAE is a video VAE compatible with existing image and video models that offers spatio-temporally compressed latent spaces, enabling efficient training of new video models using pre-trained ones.