Check this out: https://thispersondoesnotexist.com/
The site shows images of people who do not exist.
The system involves two networks: a generative network and a discriminating network.
The discriminating network is initially taught to be acceptably accurate at telling apart false data from true data.
The generative network's training objective is to become better at fooling the discriminating network by generating false data that accepted by the discriminating network.
The human faces generated by the generative network of this application are amazingly good. A few may have stranger errors such as fingers glued together. But I've never seen more realistic faces.
A typical examples: