Grounded human-object interaction video generation