Generative video models focused on people