When technology is used to make art, sometimes it produces beautiful results. And other times things get a little … strange.
A new video going around Twitter shows an eerie girl moving her eyes as a cursor moves around on the screen. The thing is … it’s not a real person in the video.
These beautiful and unsettling videos (okay, they’re more than a little creepy) were made by artist and researcher Branislav Ulicny, who created this AI-generated art by using neural networks combined with two other existing existing technology-based art projects.
“Virtual humans are kinda my obsession, so whenever I stumble upon some interesting data, I try to see what I can make out of it,” Ulicny said in an email. He was inspired by projects like Pickle Cat to work on a similar interactive experience.
To create this new, unsettling videos, Ulicny used the work of Michael Tyka, an artist who works with neural networks and created a series of AI-generated portraits, as the base portrait.
“It uses a technique called “generative adversarial networks” (“GAN“) where two artificial neural networks are playing an adversarial game: one (the “Generator”) tries to generate increasingly convincing output, while a the second (the “Critic”) tries to learn to distinguish real photos from generated ones. With time, the generated output becomes increasingly realistic, as both adversaries try to outwit each other,” Tyka explained via email.
In other words: two algorithms work together to improve each other and create the most realistic images possible.
“Using machine learning as an artistic tool is a fascinating and nascent field with many opportunities for experimentation,” says Tyka.
Ulicny then combined the portraits with Yaroslav Ganin’s DeepWarp, a project that uses images and produces “gaze manipulation” or eye movement. Here is an example of DeepWarp in action on a photo of Chris Pine:
Put together, the two result in an unsettling mix of art and terror:
The gaze follows the movement of your mouse on desktop or touch on mobile and is creepily accurate. You can try it out for yourself here.