The Latest Art Fad
The buzz among artists, when they buzz, is Generative AI. The last 2 years or so have seen an explosion of art that can be created through machine-learning software, much of it using prompt-based platforms like ChatGPT or Stable Diffusion.
Some of this is familiar. The art market, whether it’s Cubism or Pokémon trading cards, has always been about chasing FADS. Fads, of course, are those things that hungry artists ride like waves off of the South African coast. The hungry artist glides from wave to unstable wave. Some of them do it gracefully, some fitfully. More than a few wipe out, treading water as they realize that these be SHARK-infested waters! Many artists, though, are happiest building sandcastles on the beach, hoping to make something pretty without the agony of that dance we see out there beyond the shallows.
Recently I had a conversation with an artist friend. A master of ink drawing in the analog world, he was fretting about the rise of generative AI, and what it meant for the value of art. Economically speaking, it was easy to see where he was coming from. How many artforms–pottery, weaving, and other handicrafts–have gone from being bespoke artforms to products of faceless manufacture? Generative AI adds a particular sting because it learns from images of art done by humans, meaning that potentially everyone who shares their art with the world through the Internet has been training the corporate hive-minds that threaten to replace them.
Making art with AI
But have you ever TRIED to make art with AI? I’ve tried on two separate occasions. The first time, looking for laughs with Jerry after a long day of promoting our comic book workshops at libraries in Illinois, we tried to get ChatGPT to write something in the style of Jack Kerouac, and then William Burroughs, and then Charles Bukowski. In each case, especially with 20th century writers, the same pattern kept popping up. “The pudding is in the Utmost Mind, man.” “The cockroaches festered in the cupboards of the Bardo, man.” “It was a rotten, stale armpit stench of a day, but the beer was good, man.” All of these examples are completely invented, and each of them are more interesting than what the AI actually wrote. But it became obvious that Chat GPT, somewhere, had decided that in order to sound contemporary you had to adopt the cloying speech pattern of a Greenwich Village beatnik circa 1956, and maybe throw in a few Buddhist words if they mention Kerouac.
The second time I’d at least learned a little about how generative AI actually works. Which betrays some of the folly of the whole mess. First things first, what do you want the AI to do? This is more obvious when it comes to visual art. With writing a decent amount of grammar and spelling can fool the reader into thinking that this is at least Writing. Visual art is less about rules. Rather, the rules (perspective, anatomy, contrast) just attempts to bridge the gap between what the human hand can make and this general, intuitive sense that human beings have of something looking (dare I use it? The UNSPEAKABLE WORD?) beautiful.
Or even “something that looks like what I wanted?” The conversation with my friend revolved around the idea that people making art through generative AI were taking the easy way out. Or rather, they were trying to take the easy way out. When I tried to make AI art the second time I first typed in a description of art that a girl had done at one of our comic book workshops: Spongebob characters dressed as Mean Girls.
The girl had just seen the remake, and inspired by our love of SpongeBob she bridged the two concepts and produced little humanoid figures, with facial features reminiscent of SpongeBob, Patrick, and Squidward, dressed up with cargo pants and pink shirts. Even without knowing the exact “prompt” that she used (something that she made up all by herself while Jerry drew a wacky picture of Handsome Squidward), a human with some cultural context would identify the major markers–Squidward’s skin, Patrick’s star-points, long hair and female characteristics along with teen fashion clothes–and see what she’d meant.
I put “Spongebob characters dressed as Mean Girls,” into Stable Diffusion. What appeared were these melty-face, long haired gremlins with big googly eyes, contorting as they smiled, looking like they were taking a family picture with a grainy, 1990s camera. The prompt is not enough.
If At First You Don’t Succeed: AI Takes Practice, Too.
If you’ve ever tried to use a generative art-making AI you’ve probably seen something like this. Fundamentally, the machine doesn’t know what you want. A kid drawing spongebob knows what Spongebob looks like–if they get frustrated it’s because the result doesn’t match their internal picture of what Spongebob looks like. Their skill doesn’t match their WANT.
I tried again, and again. There are prompts as well as “negative” prompts, so you can tell the computer what you don’t want. No human faces, no human bodies. Then in the regular prompt category I added “Spongebob,” “Patrick,” “Squidward.” Throw in the word cartoon, cartoon eyes, cartoon faces. No doubt a more practiced AI user has more sophisticated methods (feel free to share, if you’re out there!). After about 15 minutes of this (more time than the girl had spent drawing her picture), the results looked a little better. There were still human female-shapes, with faces that looked like they’d been glued onto the heads in some kind of collage. A prism shape with eyes looked kind of like Spongebob, placed on a potato-shaped body with a pink prom dress clinging to it.
In computer speak, this is called iteration, running operations again and again until a desired result is obtained. The artspeak version of this is practice.
Computers can make truly stunning Generative AI art. But what you see on social media is just like that art people make with their hands: it is the product of thousands of hours, as the human prompter learns how to talk to the machine, how to feed it images that will produce the desired result, managing frustration and failure until the result looks perfect, effortless. As with professionally-drawn art, we see the result, not the process.
Pick up that Pencil! The Power of Making Your Own Art
As a comic artist, there is an immense power that comes from giving your vision over to someone else, to making something that belongs to neither one of you but arises out of dialogue between people. Whether AI ever gets to that place is less a matter of programming power and more a matter of developing empathy and understanding on both sides of the humanity gap.
But the thing about YOUR art is that you have the power to SEE it with your mind’s eye! And learning to DRAW your art makes you BETTER at imagining it!
A few months ago, at a comic con, an enterprising guy asked if we would be interested in drawing art for a trading card game he was inventing. He showed us a few samples he’d made with generative AI–some interesting stuff, definitely more put-together than the melty-face girls I made in 10 minutes. I told him to try drawing his characters. “Even if you don’t get results like exactly what you want, trying to make it will show you what you think looks right, and it’ll be easier to communicate that to an artist, whether it’s us or somebody else.”
I like to think he’s at his drawing table right now, with a giant arsenal of colored pencils, drawing the most incredible creature art you’ve never seen. But anyhow, one of the things we make art for is to bridge that gap, between what we imagine and what we see with our physical eyes. Explaining can be fun, but it’s never quite the same. With AI, I got bored of explaining. I wanna show you all what Mean Girls SpongeBob looks like.
