In theory at least, a neural network with sufficient training in the right domain should be able to add the slight imperfections and subtle light/texture cues o

In theory at least, a neural network with sufficient training in the right domain should be able to add the slight imperfections and subtle light/texture cues our eyes expect to see, even to too-perfect 1990s CGI scenes. This would, however, require a network that can maintain continuity between frames so these details don’t appear to jump around.