32 upvotes
DreamX learns universal action codes from any video
What is DreamX: Video Diffusion Model is a Universal Latent Action Learner?
DreamX is a video diffusion model that extracts a compact latent representation of actions, staying consistent across different performers and appearances. It can be used to predict, plan, or generate actions in robotics, animation, and video analysis.
Who it is for
Anyone building systems that need to understand or reproduce human actions across varied bodies can use DreamX as a plug-in action encoder.