Gen-2 was created by chaining a text-to-depth-map model to the Gen-1 depth-to-video model