Neural network art is quickly becoming a category in its own right, with programmers building on each other’s work to create more and more fantastical algorithm-generated images. After one researcher successfully created hyperrealistic (and hyper creepy) images of people in a recent project, another has taken those portraits and animated their eyes.
It’s perhaps the eeriest machine learning art yet. The original portraits themselves, created by Mike Tyka using generative adversarial networks, are disturbing enough on their own. But the AI artist and researcher Branislav Ulicny, who goes by the name AlteredQualia online, added to Tyka’s portraits using a neural network technique called DeepWarp which manipulates the eyes of people in images to make them move–in this case, making them follow your cursor around the screen.
As you flip through the images of imaginary people that a computer and some complex code brought to life, it’s hard not to feel like they’re staring right at you with an all-too-human twinkle in their eyes.