GAN Art: create a work of art? Klingemann lets the AI ​​do it!

 [ Art and AI , episode 9] 

 Have you read the other episodes yet ? Find out what a GAN is ! 

As we have seen in previous episodes, the invention of GAN has been of fundamental importance for the development of Artificial Intelligence. But not only! He also brought a breath of fresh air to the world of art: a few years ago the artistic movement called GAN Art was born , of which Mario Klingemann is the pioneer.

Klingemann is a German programming enthusiast, currently employed at the Google Cultural Institute in Paris. What led Klingemann to approach GANs, as he explains himself, is their ability to generate surprise and serendipity: you can train the machine to perform something specific, but it still seems to have a will of its own that leads it not to follow every literal education.

The analogies of GAN Art with other visual arts

The artist sees similarities between GAN Art and "real" painting. The process that leads the generating network to create an image that is credible for the discriminating network (I refer you to the reading of the last episode to understand it better) is comparable to that of a painter who begins a painting, leaves it unfinished, and then collects it. a few weeks later and repaint it, while leaving intact those elements that he considered good.

Klingemann also captures similarities between GAN Art and photography, in fact he says that in the same way a photographer goes out of the house and frames what he thinks best, so the artist who works with GAN "enters" these neural networks, which are like multidimensional worlds, and asks the machine: “tell me how the image looks at this precise coordinate”.

Klingemann's work, between porn stars and stick figures

Klingemann divided his work into series of works. The 2017 “Pose-to-Picture” series is certainly among the most fascinating. To complete it, the artist developed his own GAN neural network model by training it with a data set consisting of stick figures of human postures and pornographic photos (so that the machine could better understand human poses by analyzing human bodies in their nakedness. ).

As can be seen from his works, Klingemann's strength lies in enhancing and radically transforming existing images, creating something halfway between realism and abstractionism. Klingemann's use of a poorly trained GAN enabled him to do just that: the generation of vaguely human figures, very far from realism, but certainly not abstract.

Memories of Passersby I , the work in constant motion sold for £ 40,000

In 2018 his work Memories of Passersby I was one of the first artificial intelligence works ever sold by an auction house (sold by Sotheby's for £ 40,000 on 6 March 2018). You can see it in the video below.

Memories of Passersby I is not a sculpture or a painting. It is a computer that, using a GAN, generates portraits of totally invented people in real time and displays them on two 4K screens connected to it. The computer system is hidden inside an antique-looking piece of furniture that looks like a vintage radio.

The GAN used by this work was trained with paintings representing poets drawn from European art history. The images displayed on the screens (male subjects on the left, female subjects on the right) are generated in real time, therefore they are produced, not reproduced.

Klingemann specified that the images produced by the software are not the works of art, but the computer code that generates them.

Klingemann today, art and music

Klingemann's visual GAN ​​Art, in 2020, appears to be exploring musical territory. The artist has in fact collaborated with the great Massive Attack (remember that masterpiece of album that was Mezzanine ?) To make several videos of their latest EP entitled Eutopia .

The GAN Art article : creating a work of art? Klingemann lets the AI ​​do it! comes from TechCuE .