VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models

Open in new window