Stochastic Video Diffusion Models

摘要

We propose a framework for training diffusion models on video data. Our approach builds on recent advances in diffusion model training and applies them to the challenging domain of video generation. We demonstrate strong results on unconditional and conditional video generation tasks.