When will a generative model produce a realistic 2-minute video, at a 256 × 256 resolution or better?