It’s good. It might have fooled me if I was not looking for differences. I find the lips do not quite sync. I expect the next level will be to take the speech sound and determine what the face must be doing on the inside to make these sounds. That will happen. Probably soon.
Where will this end? I don’t think there will be an arms race in watermarking and encryption. At some point there are going to be so many easily-made convincing fakes out there…
In the end we will have to work out what to believe for ourselves by cross-checking, which is sort-of what we did (or failed to do) with newsprint and radio.