A model that predicts future scene states sequentially, one timestep at a time, based on previous states.