Synthesia updates – April 2022 (The new generation of avatars)

That's pretty good. These avatars are now managing to climb almost all the way out of the uncanny valley. In particular I found the more dynamic body language of the Maia avatar disarming.

Actually articulating what's still "wrong" is getting harder (but here goes):

The head movements are too smooth, with fairly even "easing" between positions. The speed should vary a bit, people aren't that consistent.

The head movements are divorced from the speech. People indicate emphasis, however slight, with the timing and magnitude of their movements in concert with their words. The nods they've added help, but those are only between sentences.

The head movements seem limited to pointing the face in slightly different directions (ie. pitch and yaw). There doesn't seem to be much in the way of left/right tilting, or shifting forward/back or side-to-side.

Overall, the current head movements convey a somewhat floaty or dreamy impression, though again, the nods do help.

With the exception of the Maia avatar, the postures are too consistently upright and neutral. There is movement, so the avatars don't look stiff, but no one leans forward at all, or shifts one or both shoulders back, or shifts their weight from one leg to the other. It's faintly menacing, as it's my experience that only highly skilled martial artists are that consistent in maintaining themselves in a relaxed and balanced posture as a matter of course (as opposed to doing so for a few moments as an affectation).