Using advanced algorithms, artists like Sofia Crespo or Robbie Barrat revisit their artistic creations using artificial intelligence. A revolution ?
“It’s like a ballet between man and machine“, says Jason Bailey, collector and one of the most famous bloggers in crypto-art. For the most part, these digital artists work with supercomputers and programs known as generative adversarial networks. They are two algorithms artificial intelligence, called “neural networks”, which compete to deliver the most complete image to the artist: the human first provides source images and adjusts the parameters to obtain, from these here, a result that interests him.
Sofia Crespo, 30, uses it to recreate animals. The objective “is not to avoid true nature, but to generate a contact with nature in a medium in which we spend a lot of time, which is the digital medium“, she told AFP in a video interview from Lisbon. Her insects are strangely hyper-realistic, with antennae, wings and bodies that seem to come out of an entomology textbook. Except that they they are all missing a head, and their bodies appear to have undergone multiple genetic mutations.
The dazzling progress of AI suggests a world where the computer would be able to learn and create, like humans. But, for the moment, she still needs to be guided. Sofia Crespo’s series of insects thus required countless round trips between the models proposed by the artist and the neural networks. Whether “the computer is an integral part of the creative process“, “the ability to generate realistic images does not make everyone an artist“, a quality that supposes “a critical and innovative capacity“, according to Camille Lenglois, curator at the Center Pompidou in Paris.
Robbie Barrat started his work around 2018, as Sofia Crespo. He entered thousands of classic art nudes into his computer, and began a dialogue with the machine, until he got what he was looking for: a series of amorphous busts, halfway between Salvador Dalí and Francis Bacon.
“When I work this way, I don’t create an image. I am creating a system that can recreate images. Somehow I create a tool“, he says. One of his works, NudePortrait#7Frame#64sold in March at Sotheby’s for more than 700,000 euros.
A lawyer’s chair?
But today, new simpler algorithms, called “transformers”, are preparing to upset this fledgling universe. One problem so far was “input text and output images“, explain to AFP Hugo Caselles-Dupré and Gauthier Vernier, from the Obvious collective.
In particular, it is necessary to be able to make the machine ingest astronomical quantities of images of all kinds accompanied by descriptions. A titanic task only accessible for economically well-off projects, such as the Dall-E 2 model from the Californian start-up OpenAI, notably funded by Elon Musk, or Imagen, a competing project from Google Research.
From a simple sentence, the machine then becomes able to mix concepts and create several representations of a “radish couple skateboarding“, of a “lawyer’s chair“, or a “astronaut monkey“, all with a photographic, comic book style, or in the manner of a 17th century Flemish painter.”This is the best in terms of image generation in general“, believes Sofia Crespo, who was able to experience Dall-E.
Experts believe that these programs could revolutionize the entire image creation and editing industry. On social networks, many examples are already circulating, including a disturbing representation of an animal at the crossroads of shrimp and centaur, from a research laboratory called Midjourney. On the other hand, OpenAI and Google have not yet published or marketed a consumer tool directly, in particular because of the risk of malicious use.