My Month with Midjourney
In my one-month experiment with the AI art platform Midjourney, I generated 12,142 images (using 183.65 hours of processing time), published 667 posts to a dedicated Instagram account where I amassed more than 300 followers, and felt, well, all sorts of ways.
It’s clear that an upheaval is at hand. Stunning artistic images are just a few finger taps away, available to anyone willing to pay the entry fee. (My subscription, which ran from November 23 to December 23, 2022 was $30.)
Seemingly, the entire visual catalog of human art and achievement — which we’ve spent the last few decades digitizing and uploading to the externalized collective consciousness of the web — is served up for plunder: painting, architecture, ceramics, movies, TV shows, video games, advertisements, products, you name it.
The arrival of AI feels like a printing press moment, a mechanical reproduction moment, an analog-to-digital moment. It’s DNA sequencing, oral contraceptives, jet engines. I’m fascinated by and horrified about how AI will likely disrupt the spheres of military planning, national politics, espionage, commerce, and so forth.
It’s also sure to upend the arts and creative professional jobs across industries from academia to cybersecurity to health care.
Photographer Shane Balkowitsch, known for his stunning wet plate portraits, recently shared an article with his Facebook followers on how AI will eliminate the need for costly real-world photoshoots.
“I predicted this some time ago!” Balkowitsch wrote. “The new A.I. photo generators are going to do to digital photography what digital did to analog. A.I. is going to remove the need for a photographer completely. This software also does not require a set, props, wardrobe, make-up, camera, lighting or even models. How are real human models going to compete with the perfection of A.I.?”
Much ink has already been spilled on the disruptive nature of AI technology on the arts, and I predict a bumper crop of handwringing editorials in the year to come.
My project was more personal than philosophical or analytical in nature — to spend a month making AI art. To understand it as an observational participant. To experience it firsthand and report back.
Here’s what I found.
Midjourney is a dazzling, dizzying, exciting tool to play with. I described it to one friend (who later bought their own unlimited subscription) as a “slot machine for poets.”
You type in a phrase, a thought, a lyric, a stanza, an idea, a throwaway line and poof: art.
Its draw is the power of getting something for nothing: manna from heaven, a sword from a stone, an embiggening mushroom from a curious brick. It is the dog’s miracle of creating food by vomiting up what it has already eaten — with thousands of years of human creative endeavor as the feedstock.
And like a slot machine, Midjourney comes with the endless promise of a jackpot just around next corner — whether in the form of a reroll (a quad of new images from the same prompt), a remix (a reroll with some manual prompt tweaks), a variation (a quad of slightly different versions of a single image) or an inaugural excursion with an untested prompt.
For much of my Midjourney journey, I was in a perpetual state of waiting. Waiting for my various prompts and rerolls and variations and upscales to complete (taking anywhere from a couple of minutes to half an hour) so that I could start another. The program lets you have up to 10 jobs in flight simultaneously. And until one resolves, you can’t start another. For me, this created a hamster wheel of anticipation and response.
And while I was waiting for the next chance to pull the lever, I was consciously aware of how similar the dopamine hit was to a real slot machine, or that of its cousin, the social media reaction. Posting my AI creations to the Instagram account I set up for the experiment offered a second stream of happy chemicals, a complementary one-two punch.
AI art, I found, scratched an itch I didn’t even know I had. It felt like an immersive video game, but with a tangible output.
“Midjourney is the best drug I’ve ever tried,” I told one friend about a week into my experiment.
Not only did I mean the constant, rapid-fire dopamine hits, but more so that they were tied to tiny, almost effortless acts of creation — essentially an never-ending tweet-to-masterpiece pipeline. AI offers a tantalizing shortcut to the shivery frisson of creation compared to laboring for hours or weeks over an editorial article or a poem or a ceramic pot at the local art center.
Using Midjourney, I made: fake high-fashion photographs, avant garde sneakers, glitchy pixelart cities, fungal wedding cakes, forgotten temples to lost alien gods, stormtrooper families in the style of Norman Rockwell, wooden busts of Frank Zappa, cyborg hedonists crafted from melting ice cream, missing pages from the Voynich manuscript, hamburgers encased in clear jelly bubbles, Frank Gehry-inspired garlic, kirigami orreries, Japanese whiskey bottles shaped like mecha-robots, 1970s iPhone ads, weird museum exhibits of glaciers. On and on. More than 12,000 images — some 400 a day.
I was limited only by my imagination, by the breadth and depth of my cultural knowledge, by my ability to cherry-pick choice fodder from others’ prompts in the members gallery, and by my stamina. It’s a world in which just knowing the word “kirigami” opens up an entire vein to be mined — one that those who’ve never heard of kirigami have no access to. My liberal arts education served me well in this experiment, as did my four decades steeped in literature, art, science fiction and popular culture — the same training I’ve relied on in my 20-year career as a journalist, science writer and editor.
My prediction is that the ability to create secret formulae to drive unique, memorable (and thus monetizable) images will become paramount for individual artists and brands, and will define the future of AI art.
Knowledge of contemporary and historical artists, art movements and cultural references to draw upon (or exploit, depending on your perspective) in prompt-making will become currency, along with individual facility at tickling the algorithms in selective and arcane ways — like the Instagram user Art-O-Maton, whose nuanced and baroque prompts have given rise to a stunning retrofuturist World’s Fair.
Searching the public gallery, I was able to sleuth out Art-O-Maton’s public Midjourney account and snoop on their prompts’ secret sauce. As this AI art space gets more competitive, it seems only natural there will be be significant monetary value in keeping prompt recipes private. (Midjourney currently charges an extra $20 a month — nearly 70% of the cost of a subscription — to be able to create in private mode.)
In the future, Algorithm Tickler will be a job. Probably a lucrative one. And about halfway through my Midjourney journey, I would have signed up to be one on the spot.
But one thing quickly became obvious: The technology is at its best when it’s directly exploiting the labors of human artists and the likenesses of real people. It goes without saying that these artists whose works and styles are being pilfered, whose years of toil and life energy are being commodified by anyone who can type in their name, get nothing. They are painters, illustrators, comic book artists, photographers, etc. — many of whom still depend on their ability to sell their art to make a living. This process is sampling on an unprecedented scale and to an unprecedented extent — sampling that pillages an artist’s entire catalog all at once. It’s sampling that can produce more works in an artist’s style in a day than that artist could in their entire lifetime. This also applies to those whose incomes are tied to their likenesses — actors, models, athletes — as well trademarked characters from comic books, video games and movies.
And the technology is just in its infancy.
These ugly truths are true even while some of AI art’s most interesting results come from its power to combine disparate possibilities into uncanny amalgams: e.g., “in the style of H.R. Giger and Yayoi Kusama.” As humans we know there’s potential magic in this type of syzygy — like someone saying, I wonder what it would sound like if Jimi Hendrix covered Bob Dylan’s “All Along the Watchtower.” (Luckily, we already know the answer to that one.) And with AI art, one is freed from the constraints of genre and medium and time period. You might create an image, as I did, taking influence from the French painter Jean Dubuffet and the Italian writer Italo Calvino, or see your favorite superheroes as imagined by a Flemish master. (This is also how one gets Jodorowsky’s Frasier.)
I discovered that there’s a richness of composition and expression that comes from drawing on the library of a named artist’s oeuvre (or several artists’) that far outstrips what the AI produces without a specific reference cue. And the reason seems obvious: computers can’t dream, can’t dance, can’t imagine — though they have become astute curators and collators and combiners. Without a vast library to train on, without that literal and virtual catalog of human creation, the AI itself has nothing to offer.
None of my forays managed to stump the system entirely, though I did get a few overly literal responses. For example, a prompt using the slightly misremembered title of a Norman Dubie poem, the open happens in the midst of being (itself a reference to Heidegger), unexpectedly generated tennis-related imagery.
That’s not to say that this new tool doesn’t open up some genuine artistic possibility — as did digital photography and Photoshop. With AI, for example, you can work in materials hitherto unknown, such as “felted wood” — a typo for “felted wool” that I let stand.
Meanwhile, accidentally typing the command “info” (which provides usage statistics and queued jobs) into the image-generating “imagine” prompt offered a small peek behind the curtain. With nothing but this single word to go on, Midjourney coughed up generic, white, commercially pretty and vaguely science-fiction-inspired female faces. It’s clear that the program, and the technologies it relies on, have inherent biases.
In this way, Midjourney suffers from the same cultural and commercial prejudices that many have pointed out are baked into our search engines and Western culture more broadly. Scrolling the public gallery, I saw countless images where a prompt that started “beautiful model” or “beautiful woman” yielded the face of a conventionally attractive (one might say stereotypic) white, American in her late teens to early twenties. AI serves dominant-culture myopia back to us. Our racism, our sexism, our classism, our consumerism-warped beauty standards, our capitalist blindness to the living world.
My eventual strategy to counter the overwhelming whiteness of my results was to include dates and places to anchor many of my prompts, or explicitly to insert racial or ethnic descriptors to get a broader diversity of outputs. Only occasionally did Midjourney surprise me with ethnically diverse subjects without additional prodding.
Taking a step back, when you get down to the heart of our unease around AI art, one nagging question seems to be: But is it art? (And thus due some affordances that purely commercial activity is not.)
Spending countless hours jacked into the matrix of AI artmaking, and in discussions with artist friends aghast at the tsunami of AI images suddenly flooding their social media feeds, I confronted some wobbly questions. What makes something art? What is art for? How do we best draw lines between what is and isn’t real art? Can we? Should we even try?
I’m not sure I came up with any satisfying answers. Certainly no eloquent ones. For that, I’d direct you to artist Dave McKean, as an example of someone who has given a lot of thought to the subject. (A big thank you to my artist friend Ryan Stiner for pointing me to the video.)
Ultimately, I came to feel that art that is all product and no process isn’t really any kind of art at all.
That is, to me, it misses the most important part of art, which is wrapped up in translating human experience and emotion and perspective into some form of expression — but which is also recursive and reflexive, influencing the maker as well. When I sit down to throw a pot — no matter how wonky or slipshod — I learn something new about the clay, about the interaction between clay and hands, about gravity and centrifugal force, about the repeating revisions that negotiate between the fuzzy idea of the pot in my mind and the evolving pot in my hands. To make art, however humble or crafty, is to be changed by the process of artmaking, each and every time — and to be changed in ways that root down into one’s humanity.
AI-generated art doesn’t do that. So that means it’s not art, right?
Still, among those 12,142 images are ones I consider beautiful, ones I consider provocative, ones I’d put on a t-shirt, ones I’d hang on the wall of my home office, ones where I felt like my prompt-making prowess brought something that hadn’t been seen before into the world. Plus, I met some kind, smart digital artists whose work I came to genuinely enjoy and admire — like @meshfolds, @beckyandthebots and @natural_futures.
I’m finished with AI art now, but as the month drew to a close, pulling the plug on my subscription wasn’t a certain thing. I was having a lot of fun, even while holding in mind the turbulence that is surely coming for working artists, perhaps for large tranches of the arts themselves.
In the end, I feel like I’m asking you to hate these images I made, because hating them is the right thing to do — as is hating the exploitative, amoral, dehumanizing technology that makes them possible. And yet part of me still wants you to look at them and like them and think they’re cool.
I guess that makes me human.