Flexible Diffusion Modeling of Long Videos

Neural Information Processing Systems 

We demonstrate improved video modeling over prior work on a number of datasets and sample temporally coherent videos over 25 minutes in length.