Imagine any picture: Sean Connery as a transvestite, the Storming of the Bastille captured by gopro, the Vikings invading New York… All you have to do is enter these few words into a search engine and the image will be generated in an instant in front of your eyes, thanks to a new kind of artificial intelligence.
Valentin Shmite, digital arts and AI teacher: “Prompt art is a revolution that is happening today in the field of art and artificial intelligence. For several months now, we have been seeing images generated from text appear on the Internet. you’re reading a hint, the hint is a piece of text that goes through a series of algorithms, a neural network, to create an image.”
Originally reserved for the geek community, these artificial intelligences have been tested by internet users to recreate fun mashups, realistic photos, entertainment, historical events, and original artwork with a creative breath and incredible power of resurrection…
September 13, 2022
In August 2022, the AI-assisted work of the Space Opera Theater even won a fine art competition in the US.
Tool Above All
In France, the debate had already begun in 2018, when the Obvious collective auctioned off an AI portrait for 432,000 euros…
But who is the author of the work?
Hugo Kasel-Dupré, one of the founders of Obvious: “We have the same comments as photographers when photography came out. Everyone told them: “This is not art, this is engineering, this is automated”, etc. But today we understand that there are better photographers than others. As with AI, some artists are better than others!”
Pierre Fautrel, member of Obvious: “Algorithms are tools and nothing more. After you have all the art part: what do you mean? What plastic rendering are you going to give your work? How does this compare to your previous work?
Since the 2010s, we have been able to create original images from a database of existing images. But recently released programs like Dall-E, Midjourney, Scrypr work with “Text to Image” i.e. they can create complex images from photo captions available online.
Valentin Schmite, digital arts and artificial intelligence teacher: “When you write the word banana, the algorithm will not just generate a banana because it has images of bananas available, but it will analyze millions of images of bananas and determine that the banana is yellow, it has a more curved shape, etc. These are 3 potential coordinates, but there could actually be millions of distinct points, and by combining those millions of points with other millions of points, we can generate images that don’t exist.
A progress that could already bring cold sweats to some creative professions: designers, illustrators, advertising journalists…
Media outlets have even used these AIs for their front pages, notably Cosmopolitan. This software has security measures in place to avoid violent, political, pornographic or deepfake content.
Already representative biases
But these AIs already include some very familiar representational biases, as Valentin Schmite explains: “If we print Nurse Image, we will have a representation of a woman, not a male nurse. If we type CEO (PDG ), a white male will usually be represented. When a wedding photo is requested, hetero-white couples will most often be presented.
Another issue promises to be acute: copyright. The use of digital technology is years ahead of the law. At the moment, most algorithms recognize the user as the author. But if we ask the Beyonce by Klimt algorithm, who owns the piece? Internet user, Beyoncé or Klimt’s heirs? The question answered by Hugo Kasel-Dupré, one of the founders of Obvious: we need datasets of billions of images and their captions, and these billions of images, there are people who own the rights to them. And in the datasets there are copyrighted images law that people don’t know are used to create other images with an algorithm.”
After 2D images, we can imagine creating 3D objects from text. We can also imagine illustrating novels, poetry and, why not, videos, short stories and even films…