Face-synthesizing algorithms often struggle with facial details like eyes and teeth. These features sometimes get pinned to a fixed spot as a head turns, resulting in an uncanny valley dweller.

A new algorithm, StyleGAN2, fixes this problem and produces "eye-poppingly detailed and correct images." It can also generate never-before-seen cars, churches, and animals.

Image: YouTube Read the rest

https://boingboing.net/2020/02/01/new-machine-learning-algorithm.html