bring in prediction of v objective, combining the findings from progressive distillation paper and imagen-video to the eventual extension of dalle2 to make-a-video