50 AI Art Fails That Are Both Horrifying And Hilarious

While AI art has become increasingly popular in recent years, it has likely been around much longer than you realize. According to TechTarget, some of the first versions of artificial intelligence art appeared in the late 1960s. However, the first notable program came about in 1973 when Harold Cohen developed Aaron. Aaron “was an AI assistant that used a symbolic AI approach to help Cohen create black-and-white art drawings,” Sean Michael Kerner at TechTarget explains. But following Aaron, it took many decades for AI art to find its way into the mainstream.

In 2014, discussions started about GANs, a foundation of generative AI technology, and in 2015, Google released DeepDream, which uses “a convolutional neural network (CNN) as an experimental approach to AI art,” Kerner says. And over the past 5 years, AI art has only become more and more prevalent, as Ganbreeder was released in 2018 (later rebranded as Artbreeder), using GAN models to allow us to use AI technology to edit existing images and even create new ones.

AI art also made headlines in 2018 after an artist going by the name of Obvious auctioned off a painting titled Edmond de Belamy, that he created using GAN models, for a whopping $432,500. “Sold in an original gilded wood frame, the portrait depicts a man dressed in a dark frock coat with a plain white collar showing underneath,” Dezeen reported at the time. “The work looks unfinished, with indistinct, blurry facial features and large areas of canvas left blank.” This was the first AI-generated piece of artwork to be sold by a major auction house, but there will likely be many more in the future.

In May 2022, Google launched its Imagen technology as a new way to create AI art, which uses text-to-image. Since then, various other programs have popped up, including Stability AI, Adobe Firefly, Midjourney, Prisma and more. And while AI art has long been a controversial topic among those in the art community, TechTarget notes that some of the ways artists can utilize AI are for art therapy, democratization, education, enhancing their existing creations, to create completely new pieces and for inspiration for new pieces.  

To gain more insight on the topic of AI art, we reached out to Angus Russell, the founder of NightCafe, an online AI art studio. Angus noted that it can be quite difficult for beginners to achieve what they’re going for when creating AI art. “A newcomer will typically write a prompt like ‘A dog riding a surfboard’. Depending on the model they're using, they may get something that looks like a dog riding a surfboard, but it most likely won't look very good because there's no information about how the image should look in general,” the founder shared. “E.g. should it look like a photo or a painting? What kind of style? What breed of dog? Etc.”

“An experienced prompter, on the other hand, would write a prompt like ‘A cute golden retriever riding a surfboard at dusk. Intricate details, HDR, beautifully shot, hyperrealistic, sharp focus, 64 megapixels, perfect composition, high contrast, cinematic, atmospheric, moody’. All these extra keywords give the AI a lot more guidance, and the keywords that experienced prompters use are generally known to be well understood by the AI,” Angus explained.

We were also curious if there are any types of art that AI is particularly skilled at creating. “It depends on the model you're using, but all models are generally very good at generating things like landscapes, lighthouses, architecture, beaches, etc.,” Angus noted. “Older models are generally not good at generating faces without making them look weird and distorted, but newer models are getting much better at that, to the point that most AI art these days seems to revolve around people and faces (often in an anime or concept art style).”

Angus shared that the best and most creative AI artwork he’s seen was an old piece, made in the days before DALL-E and Stable Diffusion were available. “The models were a lot slower and a lot less capable,” he noted. “But this AI artist managed to do something truly incredible. He sliced a photo of the Mona Lisa up into 9 rectangles, then used each of those rectangles as a ‘start image’ for 9 different generations with a post-apocalyptic theme, which caused the post-apocalyptic images to vaguely resemble the shapes and colors of the start image. Then, he stitched those images back together in such a way that if you squint, you can still make out the Mona Lisa.” If you’d like to check out the image, you can find it right here.