AI making art has been incredibly controversial. And I mean raging, frothing, screaming masses. I’ve been doing some reading and investigating, myself. Especially since deviantArt introduced an AI function (and trained it on DA’s library of 5 billion artworks without the permission of the artists). Oh, they’ve added an opt-out checkbox NOW, but yeah, they already scraped everybody’s art, whoops.
What these AI do is look at a ton of artworks, usually on Google image search. They ‘learn’ from those artworks and spit out something that looks like them. Similar, but different. It’s not just an aggregate of various artists, as so many artists claim. You use them to make art by typing in keywords, the more specific the better, and the AI tries to make something out of those keywords.
Now, in past versions of, say, Stable Diffusion, they did allow people to type in “in the style of [artist]”. This means that, like it or not, the AI really was studying these artists and making ripoffs of their work. But the developers smelled litigation in the wind and took out the option to target artists. The artists’ work is still in the AI library, it’s just harder to get at. (I’ve noticed you can still tell it “in the style of [cartoon, TV show, comic, etc.]” and it will work.)
Artists are threatened by this. “If a computer can make any art in any style, what’s the point of me spending thousands of hours developing my artistic skills?”
Well, they said Photoshop would replace physical painting, too. Instead, it became another tool that people use to communicate. That’s what art is. Communication. If you combine your refined art skills with AI, you can go interesting places.
My husband recently installed Stable Diffusion on his PC and downloaded some AIs for it that had been trained on various anime libraries. If you want to do the same thing, the files and instructions are here. But only my husband’s brick of a gaming PC could run it. My little laptop choked and died of memory errors.
Anyway, I started playing around with it. I wanted to make some character costume concepts that weren’t ripped directly from a videogame somewhere. Here’s a few concepts. I had used the prompts “handsome man, overcoat, black coat, spiky hair, armor, character sheet”.
These are some of the nicer ones that didn’t have any obvious AI glitches, like screwed up faces or too many legs. And wow, can it get hands wrong. But I just wanted some costume designs, and it delivered some pretty nice ones.
Then I decided I wanted to generate a landscape and paint over it. So I input the keywords, “Forest, red leaves, Greek ruins, steps, pillars”.
Ahah, that last one had the composition I was looking for. I like the second to last one, too, it’s a bit more moody. I might do something with that later.
Anyway, I just drew my own characters into the scene and repainted about 90% of the image.
Suddenly I have a new tool in my arsenal. I can make concept art of my books and locations. (Concept art is traditionally rough and cobbled-together anyway, just something to capture a concept or a mood.) So I’m going to keep playing with this new toy, and learning it, and I’m going to produce new stuff with it. Oh, and have a cute Jayesh my husband made for me.
K. M. Carroll, Author 