Watch This AI Paint Portraits Alongside A Real Artist

Is this machine art or human art?

Art created by neural networks has taken the internet by storm this year. But one artist is going a step further–by collaborating creatively with an AI live.


The London-based artist Patrick Tresset is training a neural network–a type of machine-learning algorithm–to interpret his portraits as he draws them in real time. As he sketches the outline of a face, slowly adding eyes, a nose, and a mouth, the algorithm draws a vague semblance of human features that correspond with the sketch. He continues to add detail and the algorithm continually morphs its corresponding drawing. The final neural network-drawn images have a soft touch to them; they’re almost like hazy, gray watercolor paintings.

Tresset trained a neural network called Pix2Pix on 21,000 image pairs that feature sketches of 3,500 different people. Pix2Pix, which was behind a drawing tool that went viral earlier this year, is a popular algorithm with artists tinkering with machine learning. You feed it any training set of drawings paired with photographs of the sketched subject, and the algorithm will generate relatively photorealistic images based on new sketches (the results are usually exceedingly creepy).

But the database of drawings used to train Pix2Pix in Tresset’s case weren’t done by hand–they were drawn by an algorithm that Tresset designed. The artist’s primary practice over the past six years has been creating and programming robots, equipped with a hand-like appendage and a pen, to sketch people live from observation. Many of Tresset’s installations have included several of these bots, sitting in a gallery and drawing anyone who sits down. Over time, they’ve sketched thousands of people, and those sketches paired with photographs of the person they drew provided the input data set on which the neural network was trained.

Neural networks are used for practical things like computer vision and self-driving cars. But they’ve also become the backbone of cutting-edge artistic and creative endeavors, like incredibly addicting drawing tools, eerily lifelike portraits of people that don’t exist, and striking stylistic experiments. Tresset’s own experiment is unique in that it’s an uncanny mix of human and AI. It’s a maze of man and machine interdependence, the two working together to create something like art.

About the author

Katharine Schwab is an associate editor at Co.Design based in New York who covers technology, design, and culture.