A Little Background
This is going to be a long post, potentially controversial too.
A week or so ago, I came across this amazing little story about the Colorado State Fair art show. Please read the following article to get a little context on what I’m about to discuss.
https://www.koaa.com/news/covering-colorado/ai-created-artwork-wins-first-place-at-colorado-state-fair-artist-receives-criticism
As someone with an art background, this left me astonished. I am still floored by this picture, and I can see many of the surreal issues of AI art in here, but at the same time the whole thing has left me with a severe case of cognitive dissonance. I went to college for art. I am a self trained graphic artist as well. I know how hard it is to actually create this kind of art “by hand”. I can’t do it myself. My focus wasn’t on fine art, it was on becoming a comic book penciller. Ah, the follies of my youth turned round again.
I was so amazed at the possible potential of this tool. It fascinated me as I tried to calculate what I could do using my own knowledge of art. This was merging with my frustration at being unable to find an artist to work with for my next set of book covers whose work aligned with my vision and that I could afford. The deadline is fast approaching after all. I was throwing up my hands, facing the possibility that like most indie authors I would be forced to go it alone and do it myself. But with this new technology, did I have the golden goose?
I threw myself headlong into research, and honestly, was getting very disappointed because I felt like I’d been tricked. Using Wombo Dream, Artbreeder and other free online software out there, I couldn’t come close to anything approaching what online tutorials claimed could be produced. (NOTE, I’m not disparaging either of those pieces of software, they just didn’t do what I expected, and I’m still trying to figure out how to use them.)
I wasn’t going to pay for the privilege of being in a beta test either, so I kept looking for software that through development and open source or freeware. I found it in Stable Diffusion. It’s out there for you to download and there are even some free UI versions that will give you limited access to the software that can still create amazing results.
But I still wasn’t getting it, even after I fought with setting it up to run on my machine and started testing out prompts. I watched videos of “better prompts”, looked up reddits and threads and blogs on how to use this, and yet, though I put in what I thought were good descriptors, it never really worked to my satisfaction. I wasn’t close. Not even a bit. But there was something brewing under the surface. I started to see glimpses of what Stable Diffusion could do that Mr. Allen used to win the Colorado State Fair art contest. He used Midjourney, but that’s just one of a few dozen different AI tools out there.
But then finally, after much frustration, I cracked it thanks to one prompt on a reddit.
I put in that prompt verbatim and was astonished with the results:
The prompt included “a modern city with rainy atmosphere and moody and cinematic lighting by alphonse mucha, simon stalenhag and darek zabrocki, cinematic and atmospheric, concept art, artstation trending on artstation”.
Needless to say, this was my first “success” and I was left almost in tears to realize what I had done. So I started picking the prompt apart. How did it make such a beautiful set of works? The next sets provided some interesting results.
These are multiple prompts that kept most of the core prompt from above. I started changing artists, a few different terms like modern to “cyberpunk” or “nanopunk”. City to rail yard. These changes are what generated the results above.
I now know that I can create professional quality artwork worthy of a book cover, as well as RPG book interior art. At least once I get the razzle-frazzit portraits and poses to give me good images and not distorted messes….
So after a few nights of going to bed at 0430 I was starting to get a feel of how to handle some things. In the process I started understanding what AI art is and isn’t. Why is this important? Just look at the news articles surrounding the article at the start of this post. People are freaking on both sides. Some are leaning towards “this is the end of all art as we know it!!!1!” frothing at the mouth. On the other end we have the “Join us and worship at the alter of the one true art-gawd of transhumanist glory and bask in it’s intelligence.”
Here are my thoughts so far after a week of messing about.
What AI Art Tools Aren’t
It’s not some sort of intelligence. It has no awareness. It’s not going to “automatically create” anything. No matter how stunningly beautiful it’s output is… AI created nothing. Nope. Not even close.
“But!” scream the professional threat inflators, sophists and panicked dim bulbs in the room. “Skynet! The mark of the beast! The Singularity!”
Settle down there, NPC kid. There is no “but” that’s really applicable because say what you want. It’s just data. Doing nothing on its own. There is no strange consciousness out there alien to all our comprehension and understanding of the world. It’s a nuclear missile sitting in the silo, waiting for it’s purpose to be activated. And that’s the key term. Like the nuclear missile, it’s purpose is powerful and on some level, sublime. It does what it does as close to as perfect as man could accomplish.
Think of CLU from “Tron: Legacy”. Oh come on it wasn’t that bad. Better than the first film to be sure.
These programs lack two things essential to all art: Imagination and the Creative Impulse.
An AI cannot really dream. It does not randomly come up with ideas. Not truly. I’m sure someone out there high on simulationist theology is going “But your brain is just a computer that never shuts off. It’s like a computer doing the same thing based on Random Number Generator or Diffusion Noise.” For the sake of this discussion that is not an apt comparison. It may be true that ideas come from some sort of crazy ether of random neural connections attaching information stored in your memory. Or maybe not. The point is, Stable Diffusion, Midjourney, Artbreeder et all have no independent impetus to create or ability. That must be 100% left up to the human at the controls. A tool just like an otter using a rock to crack open clams. There’s a more accurate comparison.
Furthermore, who creates the prompts? Who provides the image to work from? The human artist chooses the source image and the prompts to put things into the image. The AI then runs these instructions, and like a good servant/tool goes forth and performs the tasks to the best of its ability. But it has no real talent. Just like CLU, AI cannot CREATE new works of art. It can only derive them from what currently exists which was created by man.
What AI Art Currently Is
These are incredibly advanced tools based collections of huge sets of data. I write music too and my chosen DAW, Propellerhead Reason 12, has a dozen or so players and effects in there that automatically create chords and rhythms and beats with only me changing the knobs, buttons and sliders to alter LFOs, filters and number generators.
These are the exact same kind of tools. Just not advanced as something like Disco Diffusion. A CNC router with associated CADcam controls is also a highly advanced tool on par, A robot that a surgeon uses to work on hearts and brains doing work physically impossible for man to do by hand is comparable.
AI Art programs scour for and collect metadata, as well as image analysis to try and make contextual connections in which to extrapolate a new variation from the sample metadata it could find. What does the word “eye” mean? What does it look like and in what context is it being used? Now find it’s visual representation and portray that option. Do it a hundred times, and get different random results. None of them are precise duplications. They are extrapolations, making something new… but familiar.
“AI art is nothing more than a washing machine for intellectual property.”
unknown
Sorry to whomever it was who said this first, but it’s both an incorrect but very apt statement. It’s apt, because it does strike the key fear for most artists. Also in the same article, it is believed by the author that AI is no threat to any of the “big names”. Probably true, but it’s going to force them to adapt. Like when “Queen” finally gave in and discovered synthesizers and promptly went on to create some of the best songs in Rock and Roll.
Put It In the Blender & Hit Frappe
A key component to successful prompts is finding the artist who’s style fits what you want. From what I understand this is part of how they “train” the program to produce images that look like the artist… or artists. At first I was very concerned about this. I mean it was biting my soul’s keister, hard. Was I stealing from the artist? Is not their style what they own? Well from what I can figure, yes and no.
Unless intellectual property laws change significantly there’s no way to stop or protect the rights of an artist’s look. I mean how many of us have wished, or have tried to copy the masters either to learn, or to sell our own stylistic copies. Even Weird Al has a whole subline of stylistic parodies out there that are essentially original works. They sound like they might have been written by the performer in question, but they’re an original tune in the same style as the chosen artist.
Look closer and you see how genres are started. There was no “Grunge” till Nirvana, Radiohead, Mudhoney and Pearl Jam blew up the radios and the Billboard Top 40 chart in the early 1990’s. Then suddenly EVERYONE wanted in to that sound. This is where Doo Wop, Motown, Gospel, Bluegrass, Zydeco, Folk, Punk, Rap, Techno, Hip hop, Dubstep, Trap and on and on and on.
AI Art will be no different. Put multiple artists in your prompts and you get artistic mashups the same way this works.
Like students making copies in the Louvre. What if one gets bored and makes a copy of Nighthawks with cartoon characters… or like this:
There’s a whole episode of “Daria” dealing with this when Jane Lane discovers she can make money by doing copies of art that rich snobs will pay a fraction of the cost to have something that looks like the original. Surprise twist, the process of doing these copies burns Jane out on doing her OWN original art and she actually stops creating despite churning out masterpieces. Only after she quits is she able to create her own art again.
The art becomes unique but familiar, and none of the sources can claim copyright even if something looks similar. So if the law changes on this, then you’ll see radical switches in the AI art landscape. I already have a few thoughts on what will probably be attempted. But that’s a bridge to dynamite on another day.
The Sorcerer’s Apprentice
Disney once again set the perfect example, except this time it was Uncle Walt, not the current palavering pack of NPC wokies that did it. In Fantasia, the short “The Sorcerer’s Apprentice” is an incredible example of what is going on. Remember, that composer, Paul Dukas, wrote this in 1897. He had no idea visually what story could be adapted from his music. All he had was a story in his head of an apprentice who gets ahold of his master’s tools and runs amok till he he gets in over his head and the master awakes to put this right and punish him for his action.
It was Uncle Walt’s genius to make it Mickey Mouse animating brooms to carry water and fill the cistern. Think about that. The visuals alone make it a classic of modern cinema. Go watch it somewhere if you want. Knowing Disney I tempted fate by including the Tron: Legacy clip.
How did he and his writing team come up with the idea of animated brooms that then replicate from their broken pieces when apprentice Mickey can’t make them stop? That is the synergism that comes from the creative. That is something AI art cannot do without the creative impulse.
But there are a few alligators lurking in this swamp too.
Once the smoothie of artistic power is done, everything will look the same. People gravitate to the same prompts, the same artists, the same original creators and mine their style to create their own fresh take. This is where my cognitive dissonance kicks in. Everyone’s creations start looking the same, and everyone can do it now. Just like when Quark Xpress came out and the desktop publishing boom began. Technology and production FAR out of the reach of the common person suddenly came in reach. People who wanted to write news letters or magazines or little local papers but couldn’t afford professional grade production facilities suddenly could! They went from stuff done on a typewriter and a mimeograph machine to quality work on par with the biggest publishing houses in almost every respect. The synthesizer, sampler and sequencer made the home studio possible. The internet democratized publishing and all of intellectual property.
Of course, with everyone using the same filters and presets that were “industry standards”, homogeneity became endemic. Savior became enslaver. The Wynand Rule wields its cruel cudgel, for the audience will only be nudged so far to things outside it’s custom before they will destroy with passion or apathy anything that goes beyond it’s tolerance. In this, AI art will eventually doom itself. The little animated brooms and their buckets drown everyone in the blessing they were commanded to bring.
To artists who are mad at AI art possibly taking jobs: you kinda only have yourselves to blame. Just saying a painful truth. Till the world collapses and the internet falls apart, this is the future, like it or don’t. This is automation come to deal out the drudgery and hardship of menial labor, or expensive craftsmanship. Just as the moving assembly line improved manufacturing, and then was refined further with roboticization, and how 3d printing is on the cusp of replacing even that. The service industry and agricultural sectors who for decades have resisted automation because of expense are finding out they must automate to survive. Yes this will cost jobs in art on the edges of the creative fields.
But does this mean the death of art? No. It’s going to find it’s new buoancy. Water always finds its own level. That salvation for art is found in my original point of the two things AI art lacks: originality and creation. Much like the angry sorcerer finding the nearly drowning apprentice and blasting the water away by sheer force of his own will, cleaning up the mess as he goes. Art has it’s own sorcerer. He’s already here, but must first be roused before he acts.
For all those lamenting that all is lost and nobody is going to want your hand made original works, I tell you this:
YOU… are art’s salvation.
You. Making art with your own hands, seeing with your own vision, building your own unique cultural expression is the salvation from drowning in homogenized, pasteurized dreck of the familiar. You will still shock and disarm. You can still be that first play of Nirvana’s “Smells Like Teen Spirit” on the radio. That first view of “The Scream” by Edvard Munch. The shock to the system that no one could foresee.
These are things that have not been crawled over yet. These are things not digested by the AI tools. But if you are good enough, maybe you too may become the next art movement for all we plebs who may have the big dreams but lack the true talent, the auteur, the craftsmanship you possess. In this there is hope of evolving and growing beyond the smooth comforting slime.
And then there are people just like you, who have a passion but not raw talent. These are your tools to use to go forth and create greater works. Crack them open to see what makes them tick and go beyond.
AI art is not the death knell of anything, it is the next revolution.