In addition to actions, a world action model can output future video