Depending on how much time the person has to run the program, quality of the pictures the program is pulling from and how closely in skin tone, etc the two subjects are. That uncanny valley is a thing of the past.
Running a simple GANS model only takes a few minutes on my system. We also have ways to identify deepfaked stuff now. The link on the other side of this article is to an audio file though. I believe audio is quite a bit easier to detect manipulation. Just looking at the waterfall should expose some anomalies in the junctures.