I am a bit surprised how shallow the comments on this one are. Look closely, while it does generate videos of a passing similarity, they aren't "photorealistic

I am a bit surprised how shallow the comments on this one are. Look closely, while it does generate videos of a passing similarity, they aren't "photorealistic" in the slightest. They are good locally across time and space domains, but globally they are as far from realism as Doom 2 was. The only explanation for the attitude I see in this submission is that most IT people trained themselves to spot CGI by looking at local artifacts, assuming that global artifacts won't happen because stuff on the scene is reasonable. There is no "stuff on the scene" with those videos, it's just mindless vector manipulation with no underlying world model. Cars wave around, trees grow a feet from each other and behave in a way incompatible with 3D perspective. Relax, it'll require at least another AI/ML revolution (or even several) to achieve photorealism.