>>108250863Right?
It would have been trivial for someone to do a once over pass of every frame and tidy up some eyes or whatever, background, shift it over, fixing it each frame.
It wouldn't even be hard to do, tedious but not hard. Whenever I have been making animations (I've made myself, not AI) and I have been skewing things around, I've had to fix weird edge cases like this where things got crushed or altered in such undesirable ways that I just had to manually fix them or I'd be permanently pissed at how sloppy it looked.
That's one of the downsides of trying to treat a 2D asset like 3D, it requires some manual retouches to make it not look shit, especially if turning is involved. (you see this all the time with Live2D where they have that face-as-a-mask effect because the rigger / artist never took in to consideration some in between middle layer to make the illusion work right.)
The biggest issues were a few details here and there and some background stuff shitting itself whenever things moved around over them, which is still AI videogens biggest weakness.
This is one way you can be sure right now if you wanted to verify something or someone was real - ask for them to wave their fingers over the top of something detailed (especially text) for at least 10 seconds at different angles, no current AI videogen can do that consistently.