AI Art: Threat or Menace?
Answer: I wouldn't exactly worry.
I think it’s time we had a real chat about Artificial Intelligence. Namely, it can alternately be called Genuine Stupidity.
AI works via “Machine Learning”—IE, programmers feed it lots of content to mimic. It’s like the old joke, “When you steal from one source, it’s plagiarism. When you steal from many sources, it’s called RESEARCH.”
AI is just large-scale plagiarism. If you don’t believe me, look up the article by The Atlantic which discussed how Meta trained its AI off of a pirate book site. (Yes, my books were on the site. And yes, my publisher is sending them a bill.)
At the end of the day, AI is a fancy search engine, and the summaries are so reliable, I make certain to check the sources to be sure that the summary is what the sources said they were. And that the sources aren’t full of crap themselves.
Then, there’s AI art threatening artists…
Again, let’s calm down some. X’s Grok AI has been doing a decent job mimicking memes, but…
Let me show you something.
Yesterday, the trend was to have Grok reproduce something in the art style of Japan’s Studio Ghibli.
Let’s run a little test and see just how “good” it is.
Item the first. I’m going to take the cover from my novel Honor at Stake.
Okay, fine. Now, Prompt: “Studio Ghibli this cover.”
The Results are...
Where did everything else go? There’s the background. So that’s neat….
Why are parts of this in… Russian?
Why does he have glasses?
Why does she have two or three sets of legs?
Eh, at least Grok gives you two pictures.
More legs. Why is there a third character? What is going on here? And why are his eyes like that? And hers, come to think of it. Again, why is Grok writing in Russian?
Err... Dafuq?
Okay let's try another book, similar vein. Book one of the current series, Love at First Bite.
And the results are….
Wait, what?
No, what?
Also how and why?
The first one has the background, but nothing else. No T-Rex. No moon. Just the halls. And why just one character standing there, when that’s nowhere near what was on the cover?
Also, title? Author name? You know, THE EASY PART.
And the second image…
It has the dinosaur, a blonde and a redhead, but none of the atmosphere or the background. The heck…? Why?
You know what? Fine. Whatever.
Let’s try a different series in a different style.
Hell Spawn has to be easy enough to mimic, right?
Results?
Not… terrible. I guess.
One image can't get the title quite right. Or figure out where hands go. Or what hands are, I think.
The second image can't figure out that a gun doesn't work without a slide.
And “NYPY” instead of NYPD? Why is this obsessed with Russian?
Given everything thus far, I should be happy my name is spelled right.
Let's try with one of my favorite Saint Tommy NYPD covers, for Crusader.
You can see why I like it.
At least the background is correct… Book S? MYPD?
That's... so very confusing.
Also, why does it keep ADDING characters?
"Machine learning," is where AI makes it more complicated than it has to be.
While I'm at it, "Crwsadder"? For the love of…
Let’s skip ahead to Hussar. (If we did every book in order, I’ll be here all day.)
That cover seems easy enough, right?
This took a while for Grok to generate.
And the results…
Why? Just... why?
Again, the background is okay. By why "NYPB?" Is Grok Russian now? Why a cane? And why the cursed gun image? Did someone feed Grok the wrong Brandon Hererra videos?
Let's skip ahead to the last three. Light Bringer. Cover by Michael Gallagher.
I want to see what this will do with a light saber.
Again, what the heck?
The background is nice. But what happened to the title? And the subtitle? Twice?
Ugh.
The next / last two Saint Tommy covers two are by NR LaPoint.
Isn't that a nice original cover?
Add prompt and…
Now these are well done by Grok...
Until you get to the details. What happened to the Subtitles? Why are the character's hands reversed? Why are there trees?
And, yup, they botched my name. Badly.
Let’s just finish this.
This is the final Saint Tommy Novel. Cover is also by NR LaPoint.
And add some Grok prompts and …
This might be the best job by the Grok Ghibli filter.
The backgrounds aren’t bad. I’m sold. The model for the character, pretty good.
Title ... so close.
Subtitle... eh. I shouldn't expect any better by now. If it’s not highlighted in the clearest outline, it isn’t happening.
Grok has done a fair to middling job of the Saint Tommy NYPD covers.
You can clearly see how I recycled the model from Crusader. Let’s try it.
What?
No. Really. What? No title?
A name on one and not the other?
No flames on either.
Getting consistency out of this thing is like pulling teeth.
Do I have to belabor the point?
AI mimics art. It doesn’t create anything. And it’s not perfect for mimicking images.
I mean, it’s fine for memes, because the whole is what matters. But when details matter? The mimicry doesn't come up to par.
Sorry, Grok. I don't think you're replacing any work-for-hire artists anytime soon.
Illegitimi non carborundum.
Pretty much my take whenever I play with AI art. It's a really fun toy, and if used carefully, it can be a useful tool. But in one of my AI Programming courses I took many years ago, my professor said something that is as true today as it was way back when. "Creating artificial intelligence is easy. Preventing artificial stupidity is the hard part."
I sat and laughed at these. I can't get AI to do what I want, either. I use Bing AI and Grok because they're free, and Grok doesn't have as many weird restrictions as Bing. (Bing will absolutely not generate a person riding a motorcycle, apparently that term is unsafe). Also I thought it was chatgpt that had been trained specifically on Ghibli? I ought to try this kind of post with my own book covers. :-D