Arxiv paper – Video Creation by Demonstration

In this episode, we discuss Video Creation by Demonstration by Yihong Sun, Hao Zhou, Liangzhe Yuan, Jennifer J. Sun, Yandong Li, Xuhui Jia, Hartwig Adam, Bharath Hariharan, Long Zhao, Ting Liu. The paper introduces Video Creation by Demonstration, utilizing a method called 𝛿-Diffusion to generate videos that smoothly continue from a given context image, integrating actions from a demonstration video. This approach relies on self-supervised learning for future frame prediction in unlabeled videos, using implicit latent control for flexible video generation. The proposed method surpasses current baselines in both human and machine evaluations, showcasing potential for interactive world simulations.