Comment by amluto

One technique I haven’t seen mentioned would be to take advantage of the fact that the world is 3D but neural networks that make fake videos are somewhat fundamentally 2D. Making good 3D computer graphics is hard and involves a lot of lighting calculations that are very different from what neural networks do. It seems like it should be possible to detect the errors that a neural network will introduce that will make the scene unphysical, at least in cases where more is modified than just small bits of geometry like the position of someone’s lips and other facial features.