To begin with, I’m not sure I’d call myself an artist. I’ve been a professional writer — a newspaper journalist and science writer — for more than 20 years. In November 2022, I was curious about AI art and started experimenting with no thought at all about where it might lead.
The question I wanted to answer was whether this technology is capable of producing art. Nearly three years and 170,000 images later, my answer is probably a qualified yes.
Qualified because AI art has huge issues to deal with: theft of intellectual property, environmental costs, the politics of the technology’s titans — not to mention the tidal wave of generated images we’re now awash in.
While many AI artists work to develop a unique voice and style, I have tried strenuously to avoid that. If you scroll through my feed, I want it to feel like fleeting snapshots from the minds of hundreds of artists — artists who don’t exist, but who it’s easy to imagine could exist. (Though, of course, I do revisit a number of themes and personal obsessions.)
I’m almost never after a particular result. I want to be surprised. I don’t ask for specific characters or compositions or scenes. I try to push the bots to hallucinate in interesting ways.
Just as magician David Berglas’ “Berglas Effect” isn’t a single technique but a variety of techniques that can be called upon improvisationally, my approach is multifaceted.
I usually anchor my prompts with an interesting or ambiguous phrase or concept: quantum gravity, particle-wave duality, goodbye blue monday, welcome to the monkey house, the unbearable lightness of being, we lived happily after the war, etc. I draw inspiration from scientific concepts, song lyrics, book titles, lines of poetry, and so forth.
To get interesting, unexpected compositions, textures, and styles, I often reference multiple artists in my prompts — photographers, painters, sculptors, illustrators, installation artists, filmmakers. My self-imposed guardrail is to include enough of them so that the final result doesn’t feel derivative of any individual artist. With a few exceptions, when you can figure out whose work I’m referencing, that probably means that prompt isn’t a success.
I also use controlled chaos to push my work through a technique I call “feeding the ouroboros.” You feed your prompts and styles back into themselves recursively, creating a feedback loop that can have profound effects. I like to use very low levels of chaos (4-8) and weird (1-2), which slowly compounds over multiple rerolls.
To increase the visual variety of a prompt, I often re-roll it with different modifiers at the end: monochrome, blacklight, neon, ultraviolet, motion blur, closeup, panic attack, etc. (These, too, can get fed back into themselves.)
I do this until I feel like the vein has been mined out. This is usually when I have what feels like a cohesive set of images that I want to post — something that hangs together to tell a story or convey a concept. Ideally, it feels like something you’d stop and look at if you encountered it in an art gallery.
I’m actually about to bring this project to a close. I feel like I’ve pretty much learned what I wanted to learn and said what I had to say. My plan is to stop after my 3,000th post, which I’m very close to. Moving forward, I will channel that energy into more human and analog pursuits.
I also hope to write a sequel/update to the essay I wrote about my early experiences with Midjourney.
