VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models Paper • 2505.23656 • Published May 29 • 24 • 2