In September 2022, Uppsala-based artist Supercomposite took the internet by storm when she unveiled Loab, a terrifying series of images featuring a dead-eyed woman and copious amounts of gore. Anyone who’s played with image generators like Dall-E, Midjourney, or Stable Diffusion knows the very basics of AI art: one feeds verbal and visual prompts into a generator, and then lets the generator’s AI produce an image that responds to the prompt.
AI artists take the results further, though, by making manual adjustments to the image or feeding it back into the generator to spawn more images. Like any art form, AI art is equal parts intention and experimentation, with artists often keeping their prompts secret to prevent copycats—and as the internet saw with Loab, the process can produce some shocking results.
Supercomposite took to Twitter to tell the story of Loab’s emergence from the sea of data that forms an image generator’s palette, and reactions across the internet were overwhelming positive (albeit terrified). Loab might represent a turning point in the controversy over AI art, in which many critics have claimed that that art created by an AI isn’t “real.” While AI-produced images can be as vapid as any other art form, works like Loab demonstrate that in the hands of a talented artist, AI art can be just as affecting and thought-provoking as a painting or sculpture.
In an email exchange, Supercomposite shared some insights into her work, including her artistic process, the question of what separates good AI art from bad, and the role that music composition has played in her development as an artist.
What’s your process for creating an image? How much of that process is experimenting with different AI prompts, and how much of it is tweaking or adjusting the resulting images?
My process varies a lot and is constantly changing. Ever since Stable Diffusion came out as open-source, new toolings for it are being released literally every day, for doing all sorts of stuff like editing images with it, animating with it, training it with your own material and so on. And I’ve been scripting some of my own functionality as well. I could be tweaking the resulting images by hand, or with the help of further AI processing. Sometimes I like the image but I want to remove some extra limbs that people in it have, color correct, or remove JPEG artefacts that the AI put in. I am also using my own images or photographs as jumping-off points for the AI. And for video work there are a lot of considerations to take into account, for example lighting and how to interpolate smoothly. On the other hand, for the main part of the Loab story … the story I’m telling is very raw and so [those images] are not touched up.
What do you see as the main focus of your art? Are you exploring any particular questions right now? Do you notice any themes or patterns emerging in your work?
I think my art is very raw and emotional right now. A lot of it is about fears we might have or things that make us sad. I am also exploring how AI art can feel authorless. If I make something that people think is scary or affects them emotionally, did I mean to do it or was it the AI?
What was it like to see Loab go viral? Did you have any idea she would become so popular?
I had no idea Loab would blow up like it did. I got featured in PC Gamer once for some tarot cards I made, I thought that would be the most attention I ever got. I knew the Loab story was good, but wow. I think Loab just has raw emotion that cuts through to [viewers], unlike the other AI art most people have seen. A lot of AI art right now is just trying to prove that machines can draw as well as people, and stuff like that. A lot of very soulless, meaninglessly ornate art whose purpose is only to be a tech demo. And then people see Loab which is so haunting and gruesome and emotionally authentic.
What first drew you to AI art?
I’ve been doing music and sound design and then started playing around with AI on the side. In the summer of 2021 my friend showed me OpenAI Jukebox, which can generate music that usually devolves into screams and gurgles. I immediately got into image generation, starting out with Deep Daze and moving on to VQGAN+CLIP. Things have come so far in a year. AI is a lot easier to get into now for the general public, but I’m still most at home when I’m staring at walls of Python code like the “old days.”
Earlier this year, there was some controversy over an AI-generated image winning a prize at the Colorado State Fair. How do you respond to people who claim that AI art isn’t “real” art?
Everything people say about AI art now, people said about photography. Certainly not all AI art is good art. I look at [the piece that won the Colorado State Fair prize] and feel nothing. I think part of the backlash is that traditional artists feel threatened by AI economically. And those are very legitimate worries. I hope to make novel things in a novel medium, not undermine anyone. And very few people have said that Loab isn’t art, which means a lot to me.
Tell me about your music! Do you see your music as something that informs your visual art?
I am definitely working toward synthesizing AI-influenced music and AI visuals. I still prefer to produce music by hand, even if I might use AI to generate some sounds and melodies, it takes much much more time than my visual art. I built my own motion controller that acts like a MIDI arpeggiator! I wave it around in the air and it generates melodies with an algorithm I wrote. We’ll see if I ever get time to integrate that into my music … life has been full-on Loab madness recently.
This interview has been edited for clarity.
(featured image: Supercomposite)
