In the girl, the right and left eye and eyebrows are very different, looks like a mashup of different faces or a bad attempt at liquify
That's not the point, those are merely minor optimisation problems that will get better with each generation. Since the breakthrough is the speed of learning.
The application potential is genormous, even for photographers or artdirectors. It allows natural language commands like "make the hair more blond and shiny" or "a caucasian tanned look for the west, and a pale, slightly asian appearance for the east". You know, the usual indeterminate language that artdirectors speak, except from now on we can actual blame them for not making sense since the computer makes no mistakes and does exactly as you wish. (They used to be able to hide behind their artistic "vision" and blame everyone else for not understanding.)
In the not too distant future this means totally convincing actors on demand. We'd like a combination of Tom Cruise and Kevin Costner but with a more reasonable nose and a slightly nordic appearance, and you can do so after the movie was shot (or rendered!?). With a bit of luck this technology is available before Top Gun II is released...
An entirely different application is easy composites of related dna, or online shopping with clothes sampled to your own bodystyle.
Or, as is also shown, car designers can go totally wild with composites of existing designs. "Show me this but with a beamer front and ferrari back". "Now change the doors to 80s style coupé". We don't have that ease of command yet, but this is bringing such a reality much closer.
