I would not call them “incredible”. Video generation is still at gpt-1 level. It will need several orders of magnitude scaling to get to gpt-2 level. We’ve just started making very first baby steps in this direction.