"we introduce a Space-Time U-Net architecture that generates the entire temporal duration of the video at once, through a single pass in the model".

Holy fuck.