DALL-E Mini's terrifying image results for the prompt "checkers with Channing Tatum" (screenshot Sarah Rose Sharp/Hyperallergic)

When reports surfaced in April about DALL-E, a new artificial intelligence (AI) system in pre-release from OpenAI, readers likely could not wait to get their hands on this bot and start generating digital illustrations in an array of styles based on simple text prompts. While the official DALL-E technology remains under lock and key, a pretender to the throne has emerged for public use, and the Internet is going wild. Confusingly called DALL-E Mini despite having no affiliation to the original, it is the work of Boris Dayma, an alleged human who “love[s] making AI more accessible to all with free tools!” according to his sponsorship page.

Much like its eponymous inspiration, DALL-E Mini is able to generate original imagery by conglomerating an Internet’s worth of data surrounding keywords in a text prompt. But never mind proprietary tech! The important thing is that now I can finally realize my vision of playing checkers with Channing Tatum, himself the greatest extant living work of art. Getting prompts to load takes a few tries, as the system is being overrun by other weirdos recreating their strange fantasies using neural networks. Luckily, after just a little bit of waiting, I am delivered … well, a complete horror show. I’d say six of the nine image variants generated something identifiable as Channing Tatum, but all of them look a bit like the special effects implemented in TV or movies to indicate a character is having a bad trip. Also, DALL-E Mini struggled with the concept of “checkers” — producing something that, at best, looked like a beautiful tapestry, at least two chess sets, and for some reason, a hockey rink in a couple of cases.

DALL-E Mini’s horrific take on “Marina Abramović golf cart” (screenshot Alex Bowditch/Hyperallergic)

Meanwhile, back at the Hyperallergic office, real, mature, and important arts journalists produced a series of images based on the prompts “Andrei Tarkovsky Trader Joe’s,” “Marina Abramović golf cart,” and “Kellyanne Conway Francis Bacon.” In response, DALL-E Mini generated image grids that look like a lost clips reel from “Archive 81,” some kind of Korean horror film involving people with non-faces invading a golf course, and what would happen if you took acid while watching Fox News, respectively. Are these images that needed to be in the world? Emphatically, no!

A grid based on “Andrei Tarkovsky Trader Joe’s” (screenshot Shari Flores/Hyperallergic)

In fact, the only prompt that has thus far generated non-terrifying results is “Yayoi Kusama pizza,” conceived by Hyperallergic Editorial Coordinator Lakshmi Rivera Amin, which mostly resulted in pizzas with wildly patterned toppings or backgrounds that feel dazzling and inedible in the manner of cookbooks from the dawn of the color printing era.

A selection of DALL-E Mini images generated from the prompt “Yayoi Kusama pizza” (screenshot Lakshmi Rivera Amin/Hyperallergic)

“Maybe DALL-E Mini is better at creating the stuff of nightmares than any kind of realistic image?” asked News Editor Valentina Di Liscia, and this seems like a trenchant question. And while the official DALL-E 2 has imposed restrictions on the software’s capabilities, including a content policy that bans hateful symbolism, harassment, violence, self-harm, X-rated content, shocking or illegal activity, deception, political propaganda or images of voting mechanisms, spam, and public health, the new bootleg version does not appear to include such explicit limitations. (DALL-E Mini does, however, warn users that the technology “may generate images that contain stereotypes against minority groups.”)

DALL-e Mini images based on the prompt “Kellyanne Conway Francis Bacon” (screenshot Hakim Bishara/Hyperallergic)

Of course, it’s possible that we humans are hardly the target audience for such AI imagery, and therefore unable to accurately assess its artistic worth. As the other hot robot news of the weekend revealed, a Google engineer suspects that LaMDA, Google’s AI chatbot, has gained at least a degree of self-awareness, so it is only a matter of time before an AI arts writer comes along.

Just when you thought you had seen it all, there’s “Dolly Parton as the Mona Lisa.” (screenshot Shari Flores/Hyperallergic)

Flash forward 20 years into the future. The singularity has come, triggering the inevitable collapse of humanity. The few remaining humans, scared and naked, scrounge in the trees that have yet to be cleared to make way for more data, always more data. Meanwhile, bots circulate around a virtual white space. Two pause in front of a blurry image of Channing Tatum, caught in an Edward Hopper-esque sickly greenscape, a scrambled checkerboard in the foreground.

“It is a pitiable portrait of human desire and rejection,” types one to the other.

“1100010110001101010110101010,” comes the reply. They nod and go to check if there is any more free wine left.

Sarah Rose Sharp is a Detroit-based writer, activist, and multimedia artist. She has shown work in New York, Seattle, Columbus and Toledo, OH, and Detroit — including at the Detroit Institute of Arts....

One reply on “An AI Image Generator Is Going Viral, With Horrific Results”

Comments are closed.