The algorithm from Nvidia allows you to create the face of a non-existent person in a few seconds. And it looks realistic.
On February 10, Nvidia published the open source code StyleGAN – an algorithm that allows generating new images using neural networks. One of the main examples of the work of the system was the creation of realistic human faces – the company showed the results of the algorithm in December 2018.
On February 14, 2019, Uber developer Philippe Wang launched ThisPersonDoesNotExist site based on Nvidia’s open source practices and invited everyone to generate a face in a few seconds. The neural network takes into account many of the smallest details, so almost every time people turn out different. When the page is updated, a new face is generated.
Western media noted the realism of the people who created the algorithm. At the same time, periodically the system still fails, and on the generated faces you can see artifacts. For several days, StyleGAN was applied in other areas: for example, they created cats and anime characters (one work was rated by Elon Musk).
— Ryobot | りょぼっと (@_Ryobot) 12 February 2019
Anime characters generated by the neural network
The new StyleGAN code is super neat. You can cross-breed latent vectors to get a sort of style transfer effect. pic.twitter.com/TXWzcZjai9
— Gene Kogan (@genekogan) 6 February 2019
Neural network generates cats
— Janelle Shane (@JanelleCShane) 5 February 2019
In some cases, the neural network generated logos of stock sites.