I noticed that too. It also doesn't seem to always know how to map (or remove) certain things, like the hair bun on the input image, to the generated avatars on

I noticed that too. It also doesn't seem to always know how to map (or remove) certain things, like the hair bun on the input image, to the generated avatars once you get outside of the facial region.