Where's my phone?

Photographer: Mario Tama/Getty Images

New Masters: Software Has Painterly Technique Down Pat

Leonid Bershidsky is a Bloomberg View columnist. He was the founding editor of the Russian business daily Vedomosti and founded the opinion website Slon.ru.
Read More.
a | A

An iOS app called Prisma has been taking 10 post-Soviet nations by storm, reaching the No. 1 spot in app stores in record time. At first glance, the software does nothing special: It's a collection of image filters that make photos look like paintings. But the technology is far more complex and troubling: It uses artificial intelligence to completely "repaint" images in any of 20 preset manners, taking a major step toward making painting technique even more irrelevant than it has already been rendered by contemporary art.

Post-Soviet developers are good at creating AI image manipulation software. In March, Facebook acquired a Belarussian startup called Masquerade, whose self-learning algorithm allows users to alter their faces in stills or videos. The artificial intelligence in the product, however, is largely limited to fitting effects to different facial shapes -- fun to play with but, in the end, it amounts to very smart tech in the service of silliness.

Prisma is altogether more clever. It's a consumer implementation of techniques described last year in two papers -- one by Leon Gatys, Alexander Ecker and Matthias Bethge of the University of Tuebingen and the other by Chuan Li and Michael Wand of the Johannes Gutenberg University in Mainz. Both German teams used so-called convolutional neural networks -- a type of artificial intelligence that, when "trained" on a series of images, can "learn" certain stylistic features and apply them to a different image -- say, a photograph. 

Gatys, Ecker, Bethge, 2015

Until last year, only humans could separate style and technique from content in a painting. "All in all it is truly fascinating that a neural system, which is trained to perform one of the core computational tasks of biological vision, automatically learns image representations that allow the separation of image content from style," Gatys, Ecker and Bethge wrote. "Thus, our ability to abstract content from style and therefore our ability to create and enjoy art might be primarily a preeminent signature of the powerful inference capabilities of our visual system."

The algorithms do not apply filters to a photo -- they only use its content as guidance for new images produced with the features "learned" from Picasso, Van Gogh or Munch. Although the German researchers published their work, Prisma's Russian developer Alexei Moiseenkov faced a tough task: Reproducing images using the neural network requires lots of computing power. And it used to be time-consuming, as anyone who played with Google's earlier experimental code that produced hallucinatory images from photos using machine learning would know. Prisma produces the images almost instantly; the code written by Moiseenkov's four-person company runs on the hardware of another company, Servers.com, which has to double the server capacity allocated to Prisma every day: The app has been downloaded hundreds of thousands of times, and post-Soviet users have flooded the social networks with their photos recreated as paintings.

It's unclear how Moiseenkov, who created Prisma as a side project, is going to monetize his wildfire success. The app is free. Asked whether he hoped his company would be acquired, Moiseenkov replied, "Anything possible, but I think we should wait until that moment." What's more important is the effect this kind of software can have on art and, potentially, other creative pursuits.

Painterly technique has lost its importance in recent decades, giving way to the boldness or intricacy of ideas. Yet developing an individual style still requires a lot of training and practice, and the results of all this hard work often are hard to replicate. The computer analysis of brushstrokes is sometimes used to authenticate paintings. The algorithms on which Prisma is based make the entire wealth of artistic styles available to anyone, and they make it possible to develop synthetic styles that would be significantly different from any known painter's: They just need specific "training" material. 

This allows for the creation of original digital art such as this picture, made with Prisma. Yet it further devalues artistic skill, replacing it with technology to an extent that was impossible until very recently. Ingenuity, stories, the choice of content become the artist's only relevant tools.

(Corrects ownership of Servers.com in fifth paragraph of article published June 20.)

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

To contact the author of this story:
Leonid Bershidsky at lbershidsky@bloomberg.net

To contact the editor responsible for this story:
Max Berley at mberley@bloomberg.net