All the videos I’ve watched — out of technical curiosity, naturally — have had some sort of glitch that made it obvious it was fake. I think this technology will have a serious problems just mapping the facial expressions of one person onto another, since many people have their own distinct facial expressions.
The linked YouTube channel with the Putin video[1] is a good example: it looks completely unrealistic because the actor in the source video makes facial expressions Putin would never make.
In my personal opinion, I think it will take decades before this technology becomes good enough to fool humans, and probably longer before it can fool humans closely related to the subjects of the fake videos — if this ever becomes possible at all. The fundamental challenge is mapping the emotions of one person to another’s, which isn’t easily solvable. Just mapping the facial features of Putin onto SNL’s Beck Bennett isn’t going to convince anyone familiar with how Putin looks and acts.