The most powerful embedding model for video understanding.