Weirdly I don't think there's a machine learning algorithm that can actually do this. pix2pix loses cohesiveness quickly, and GANs require mostly aligned input

Weirdly I don't think there's a machine learning algorithm that can actually do this. pix2pix loses cohesiveness quickly, and GANs require mostly aligned input to make any sense of reality. BigGAN is the closest, yet even that doesn't seem to work very well for continuity.