“The emergent ability of video models to understand cause-and-effect and maintain semantic consistency.”