Text-to-A.I.-Art Rabbit Holes
Ten thousand years of torment
BY QUINN MCCLURG
There is no word count high enough—nor an audience patient enough—for me to properly express the endless agony that I have found in the derelict depths of A.I. art rabbit holes. I have made abhorrent creatures and been unmade by abhorrent creatures; I have suffered by the sterile hand of artificial intelligence and I have suffered by my own hand in artificial intelligence-related situations; I have been through the worst of times, and even the worst-er-est times—Heavens! The depths of depravity! Deliver my blue-light-blinded bleary soul’s eyes!
My torment began over one year ago. I was huddled in my dorm, listless with the malaise that comes with not-busy Friday nights, when—Hark! Lo! Behold! I lighted my eyes upon—
Ok, I’m done with this charade; I thought writing like H.P. Lovecraft would make this article easier to write; however, it has not. Previously, A.I. generated art was something I would occasionally dabble with on NightCafe.com, but I woke up on some morning in mid-October, wresting with some half-forgotten dream with some fully beautiful imagery. I don’t remember exactly what the dream was (my dream journal is both very descriptive and very sparse), but I remember thinking that maybe, somehow, someway, if I took faint aspects of my dream and put them into an A.I. text-to-image generator, I might be able to remember the entirety of the dream. Thus, I began digging a hole and covering my ankles in soil.
It warrants stating that I have always been very invested in dreams; I cherish any glances I can get of that “other” that the waking, “reasonable” consciousness suppresses. I view the internet as a collective unconscious; having a program that randomly skims bits of meaning, symbols, and archetypes from the top of this pool (when given an abstract concept, emotion, or sentence) might as well be a dream, a gaze into the unconscious collective, a random collection of nothings which means nothing, but might resemble for of something the deeper you look into it. I wanted to find the shapes of our souls, the color of our emotions, the named form of the intangible-unintelligible. I didn’t even think to ask myself if this pursuit would amount to anything meaningful or profound, I just hoped it would; I believed it had to (in hindsight, someone should’ve told me to go outside).
I picked NightCafe and StabilityAI’s Dream Studio as my main poison and made as many different accounts as I could. NightCafe gives you 5 credits a day, and you can get more from accomplishing specific goals within the website’s community; however, it paled in comparison to Dream Studio as, on Dream Studio, a user is given 200 credits with the creation of an account. As with both websites, 1 credit will give you a 512-pixel by 512-pixel image, but Dream Studio grants you many more customization and manipulation options than NightCafe, especially ones that are free of additional charges. The one-credit images resulting from Dream Studio have significantly higher quality, coherence, and attention to detail than even the most advanced multi-credit settings on NightCafe. For the purposes of my brief spiral, I ended up utilizing both websites.
Over the course of three days, I generated over 700 pictures from phrases such as “a wish for stars,” “a being made of smoke and sunlight,” “I dream subterranean,” “the slow regard of silent things,” “the embodiment of rot, captured in woodgrain,” “we are blessed with such maddening lacks,” “the saintly self and spiritual starvation,” and, who could forget, “an ouroboros with a golden halo in the hall of the centipede king, a time beyond time,” among many others, but nothing was as random, abstract, or profound as I desired it to be. I adapted to adopting imagery of Cernunnos and other pagans gods, of cathedrals and spires, of caves and dreams, of sunlight, of static, of forgotten wonder, and still ended up dissatisfied. All-in-all, I ended up downloading only 35 generated images, ones I somewhat liked; a measly 5% of all the images generated. Nothing scratched that unsettling, esoteric, and distorted aesthetic that I truly sought, as all my results seemed superficial, perhaps lacking the depth that a human hand might give them.
Although I didn’t spend any money on this bizarre escapade, I did spend a lot of time—time that came at the expense of homework, hobbies, and health. I was generating images during class, when I was supposed to be sleeping, and when I was supposed to be hanging out with friends. I do feel as if this model of obsessive behavior says more about me than the average person; however, I do feel as if that pull toward the profound should be noted. When we conceptualize something as more important and more profound than us, there’s an inexplicable urge to chase it and make it tangible, especially if the subject-matter is already intangible.
So, what can be learned from A.I. generated art, my self-neglect, and our capacities for wonder? Maybe we can never achieve these impossibilities that we seek; maybe, when we try to grasp the spirits we chase, we’ll always just end up disappointed, holding handfuls of dew; maybe trying to stare at the sun is dumb and all you’ll get is blind. But as long as there are spectres to chase and celestial bodies to seek, I’m ok with ending up with soggy clothes and sun-spotted eyes, and you should be too. As long as there are wonders to chase and mysteries to fall into, there is magic in the world. Just don’t stare at the sun too long, kid.