These people do not actually exist. They were generated entirely by a machine—a specialized form of Generative Adversarial Network (GAN) called the StyleGAN. A StyleGAN uses the method of latent space. It starts by generating a small image with low resolution and then improves the resolution more and more. With each improvement, all the details are also reworked. This makes it possible to create images that seem deceptively real and cannot be distinguished from photos of real people.
Credits: A Style-Based Generator Architecture for Generative Adversarial Networks (StyleGAN) paper by Tero Karras, Samuli Laine, Timo Aila
With support by Nvidia