Art with artificial intelligence: Create your own astronaut in the style of Rembrandt

In the novel 1984 by George Orwell, the main character Winston hears a woman singing a song through an open window. It is a song made by a machine that Orwell calls “the versifier”. There was no human hand involved. Ugly music, says Orwell. But the woman sings it so beautifully that it is almost bearable. Reluctantly, he admits it. That song is hope in Orwell’s dystopia. The singing woman is not given a name by the author, she is old, fat and ugly. Is she also an artist? Orwell thinks not. I still think about it.

Because this woman—let’s call her Eve, after the first woman in the Bible and robot named after the first woman in Pixar’s cartoon Wall-E—came to mind when I first opened the computer program Dall-E (named after the artist Dali ). Dall-E is a miracle in the form of a computer program that can convert text to images. For example, press ‘astronaut on a horse’ and an astronaut on a horse will actually appear after a few seconds. You can also specify the genre: should it be a drawn, painted or photographed astronaut on a horse? And within that, countless variations are possible: painted in the style of Van Gogh or Rembrandt and much more. The possibilities are staggering.

Alongside Midjourney and Stable Diffusion, Dall-E is one of the artificial intelligence art generators that started making waves last year. They have a huge reservoir of images collected on the Internet, which an algorithm can combine in all sorts of ways. Here she looks like the chatbot ChatGPT, an equally incredible program that can brew a readable text with the help of AI from a few words. “Write an essay on Dall-E, artificial intelligence, Vermeer and lemons,” I asked the chatbot because I was busy making still lifes with lemons on Dall-E in the style of various artists. Dall-E made this suggestion: “With Dall-E it is possible to generate images in the style of Vermeer by providing a text description of the desired image and selecting Vermeer as the desired style. For example, a description such as “a lemon on a table with light coming in through a window” is used to generate a Vermeer-style image, with the lemon and table rendered in very realistic detail and the light and shadows from the scene captured with the same level of skill as in one of Vermeer’s own paintings .”

Voila. Boring but true. Probably.

Varying the command always produced a slightly different text, although the robot sometimes took my messages very literally. The request for a lyrical essay resulted in a poem, unsolicited rhyme. Dall-E also sometimes has difficulty with questions that seem simple to people. The chatbot appeared when I asked for a creative essay with Vermeer and lemons with the following suggestion: “An example of DALL-E’s creativity is the ability to generate a unique image of a lemon with a human face, complete with eyes, a nose and a mouth, all formed from the texture and shape of the lemon itself.This type of creative output is made possible by DALL-E’s advanced engine learning algorithms, which is able to analyze and process huge amounts of data to generate new and imaginative ideas. In the case of the lemon with a human face, DALL-E was able to draw on his knowledge of human facial features and his understanding of the properties of lemons to create an image that is both visually striking and highly imaginative.”

Naturally, I put it to the test. But with Dall-E, generating such a lemon head was much more difficult than the chatbot made it out to be. A lemon with a human face was above all a lemon next to a human face. It wasn’t until I added the word surrealism that it started to look like something. And it wasn’t until I let the lemon take a Magritte-style selfie that a lemon really came with a human face. Charles Philipon probably did it faster and better in the 19th century, when his caricature of King Louis Philippe turned into a light bulb. But he could draw. Not me.

Bianca Rises, ‘Without title’. (AI art created by Bianca Stigter using Dall-E)
Photo Bianca Stigter

Art for lemons

But is it art?

This is the key question for many who are confronted with artificial intelligence.

So what?

If the versifier produces a beautiful song, does it matter that it was made by machine? Perhaps a new kind of artist will emerge, one who is particularly good at writing prompts, the commands for AI. My fumbling with the lemons shows that it’s not that easy. When the American artist Karen X. Cheng for Cosmopolitan created the first AI-powered magazine cover in June 2022, it boasted that it took only 20 seconds to achieve this result. But in reality, the artist, Karen X. Cheng, had spent days writing Dall-E the right prompt. “Hours and hours of generating and refining prompts until you get the perfect image,” says Cheng on LinkedIn. ‘A female astronaut walking on mars digital art’ and ‘an astronaut with a feminine body walking on mars digital art’ did not return the requested image. It just hit the spot with: ‘A wide angle shot from below of a female astronaut with an athletic feminine body walking with swagger towards camera on mars in an infinite universe, synthwave, digital art’.

Tertiary Surrealism

A kind of cheap, third-rate surrealism is the highest result for laymen. A toy that gets bored quickly. How many seals à la the girl with a pearl earring and surf mouse can you handle? The website encourages that sense of fun by featuring things like an avocado-shaped armchair or a Van Gogh-style painting of an American football player. The algorithm had a harder time dealing with the surreal saying “as beautiful as the chance meeting between a sewing machine and an umbrella on the cutting board”. First of all, the cutting board did not pass Dall-E’s vote. The word didn’t fit into their ‘safe system’.

Can’t wait for the first AI masterpieces

I was surprised they did, because racism and sexism seem to be ingrained in the algorithms. White men sit in front of Dall-E. Unless you explicitly ask otherwise, even a lemon is a white man. Fat old women à la Orwell’s Eve are out of the question. The Lensa app, mainly used to create user avatars in familiar styles from popular culture – a way out Star wars or Game of Thrones – unbidden made me younger and gave me breasts so protruding they could be used as weapons. American artist Cindy Sherman managed to create horrific portraits using – presumably Lensa – ugly in every possible way. Orwell’s Eve would be delighted.

The American Jason Allen worked for months with AI-Midjourney, which resulted in the artwork including Theater D’opera Spatial, which he had printed on canvas and entered in the Colorado State Fair’s fine arts competition in the digitally manipulated photography category. Allen won first prize. It is not clear whether all jurors were aware of the creative process. Regardless, many critics felt Allen cheated: “It’s like running a marathon in a Lamborghini.” Others protested: why ride in a carriage when there are cars? A middle ground between all these racetrack comparisons: people still ride horses. They also continued to create paintings after the invention of photography and film. And what beautiful paintings have not produced.

Bianca Rises, ‘Without title’. (AI art created by Bianca Stigter using Dall-E)
Photo Bianca Stigter

Can’t wait for the first AI masterpieces. So far, they are often works in the style of others, not only by old masters like Van Gogh and Rembrandt, but also modern masters of digital art like Greg Rutkowski, whose work is mentioned mostly in prompts (about copyright and payments nothing has been mentioned .arranged yet with the AI ​​generators).

Forerunners

AI art now also has a long history of its own. Among the forerunners is the British mathematician Ada Lovelace, who as early as 1842 imagined a machine that could not only calculate, but also solve all kinds of problems. She called it “poetic science.” The first real artificial intelligence was created in 1973 by the American scientist and painter Harold Cohen, who let the AARON program independently generate abstract works of art. In 2015, Alex Mordvintsev developed the program Deep dream for Google, which itself understood visual concepts and was able to generate hybrid images based on a huge data set, such as a fishing dog and a camel. In 2018, the French art collective Obvious fed an algorithm with 15,000 portraits taken from WikiArt, upon which it created its own painted portrait. Portrait of Edmond Belamy (which never existed) was printed on canvas and sold at auction at Christie’s for nearly half a million dollars. The signature at the bottom right of the painting is part of the algorithm used to create the painting.

Therefore, not all artists use the datasets from large companies such as Google or OpenAI, the company behind Dall-E and ChatGPT, which still suggests the programs as a tool. MOMA in New York is showing a work by Turkish-American artist Refik Anadol, who fed images of 138,000 works of art from the museum’s collection to a computer program that now generates its own images based on that collection. “What would a machine dream of after seeing the collection of the Museum of Modern Art?Anadol wondered. He titled the work Without supervision. The machine is therefore more than a tool there. Is this machine very different from the 1984 versifier? Eva was then still needed to sing it. Soon we will only have to listen. Or to see. Or maybe not even that.

Leave a Comment