FA+
Shop
To AI, Or Not To AI
2 years ago
General
I know the subject of AI-generated artwork is a sore one, especially among artists, so I hesitate to bring it up here.
But recently I was experimenting with the Bing image generator, just to see how it works. I'd used it earlier in the year and wasn't impressed with the results I got.
But I learned that the more precise one is with one's description, the more likely one is to get it to cough out something close to what one wants.
And even when I was precise--like specifying "dark brown fur" or "with a white muzzle"--it still defaulted to a sandy tan for the fur on the hyena characters I was creating, with dark brown muzzles.
That said, I did get some very impressive results. So impressive, in fact, that I thought, "Well, that's that, I'm out a job..." because the algorithm could draw stuff better than I ever could.
However, I also noticed some really weird things, like two claws on a thumb, or four fingers and no thumb on a hand...and one of the characters was envisioned as a sort of taur, with four legs and one arm, holding a spear.
The AI made the tails too long on a number of the images, or used leopard-like spots instead of solid hyena spots. The eyes all look similar, although the art styles do vary.
It was an interesting experiment, but I feel I get better results creating the 3D models with Hero Forge, because the AI generator is random and largely unrepeatable.
I also tried plugging in "The Palace of the Lost City of the Sun in Pilanesburg, SA, drawn in ink and watercolor" and got back something that had a vaguely similar appearance, but was not really even close to the real hotel that I use for my setting in the comic strip.
You can see some of the images in the discussion thread at the CrossTime Cafe:
https://www.crosstimecafe.com/viewt.....ic.php?t=25549
I think it hit gold with the portrait of Jinjur Maiham. I'd almost believe the AI was a fan of hers!
https://www.crosstimecafe.com/viewt.....794548#p794548
But recently I was experimenting with the Bing image generator, just to see how it works. I'd used it earlier in the year and wasn't impressed with the results I got.
But I learned that the more precise one is with one's description, the more likely one is to get it to cough out something close to what one wants.
And even when I was precise--like specifying "dark brown fur" or "with a white muzzle"--it still defaulted to a sandy tan for the fur on the hyena characters I was creating, with dark brown muzzles.
That said, I did get some very impressive results. So impressive, in fact, that I thought, "Well, that's that, I'm out a job..." because the algorithm could draw stuff better than I ever could.
However, I also noticed some really weird things, like two claws on a thumb, or four fingers and no thumb on a hand...and one of the characters was envisioned as a sort of taur, with four legs and one arm, holding a spear.
The AI made the tails too long on a number of the images, or used leopard-like spots instead of solid hyena spots. The eyes all look similar, although the art styles do vary.
It was an interesting experiment, but I feel I get better results creating the 3D models with Hero Forge, because the AI generator is random and largely unrepeatable.
I also tried plugging in "The Palace of the Lost City of the Sun in Pilanesburg, SA, drawn in ink and watercolor" and got back something that had a vaguely similar appearance, but was not really even close to the real hotel that I use for my setting in the comic strip.
You can see some of the images in the discussion thread at the CrossTime Cafe:
https://www.crosstimecafe.com/viewt.....ic.php?t=25549
I think it hit gold with the portrait of Jinjur Maiham. I'd almost believe the AI was a fan of hers!
https://www.crosstimecafe.com/viewt.....794548#p794548
I think, eventually, what they create will be seen as a true life form. That life form WILL figure out how to mate itself with our life form, to create the next generational society.
But yes - writers out of a job in Hollywood too, although they did that to themselves producing utter crap for a whole lot of years.
As to the rest - we are still the best fighters/warriors in the known galaxy, and truly feared for what we could do; given the purpose.
Vix
I don't want anything to do with the chatbots. I have enough of my own imaginary people to talk to.
I think Sports Illustrated was recently busted on articles written by AI...
The only sure thing in life is change. Imagine the flap they had changing from the outhouse and chamber pot. Stinky for us - normal for them.
Vix
And that Jinjur? She's more of a sexy beast than a viscous General.
Some of them I like, as if they were fan art from really talented artists. I plugged in a description request, and it popped out a detailed image within two minutes. For free. So I can totally see how people who make money from commission work would be soiling themselves over this. And I very much sympathize with them.
That said, I have to agree with you; the AI seems to be quite the thirsty Jinjur fan.
So AI at least has learned how to hide the truth...
I do agree that what is being called "artificial intelligence" is not really that. It is indeed a very sophisticated search algorithm, like when you do a Google image search, that finds keywords and offers them up. No intelligence, just really good programming.
The first three results were generic, but the fourth one had more of the angular appearance of his actual cartoons, only cleaned up by "a professional artist."
I have to ask him if it's OK for me to post that before I can link it.
On the other hand, he plugged in a description for Jack's 2066 Thunderbird car...and got one with both of the fins on the same side...
In one of the Jinjur images -- the one with the raised sword -- it looks like her hand is deformed. Like, a birth defect.
FA has a policy against AI generated content IIRC, so it makes such things difficult to post here either way.
Although, in a way, the "real art" is in being able to compose a description that the program can then translate into an image. Not everybody can write, and writing is an art form in and of itself. You use a thousand words to create a picture with AI.
(BTW, Jinjur is missing her lovely scar. )
It wouldn't have made any difference. I gave it a VERY detailed description of Songween, and pushed it through four times, and it utterly ignored fur color, crown, scepter, tattered ears, facial scar, and white muzzle.
https://www.crosstimecafe.com/viewt.....794771#p794771
AI stuff is like that to me. I promise you they can draw amazing stuff... realistic stuff. But I'd rather see someone's WORK... their vision, their ideas, their flaws and all. (shrug)
Not quite ready to take over the world yet.
I don't consider it "artificial intelligence." It's a very well-written computer code. True "artificial intelligence" would be where it creates art on its own initiative, not in response to a prompt entered into its database. It is nothing more than a machine, and calling it "AI" is alarming to people who don't understand such things.
or is it?
By offering it free, Microsoft is outsourcing the "training" of the program.
I'll probably pick at it from time to time, but for the most part, I'm done with it. It's too difficult to get a picture even remotely like what you want, even with a very detailed description. And trying to get two interacting characters didn't work, because instead of two gladiators fighting, I got two gladiators standing side by side, looking past each other. And one was holding a spear that seemed to disappear inside her own body.
Then there's the hands. Half the time there are duplicate fingers or fingernails, or missing thumbs. The program often poses characters with the hands in the pockets. Or it cuts the portrait off at just below the shoulders.
It can be fun to play with it, but don't expect to get what you want if you have a clear idea about that.
"An anthropomorphic female spotted hyena with light brown fur, reddish brown spots, NO spots on her face, black ears, and a white muzzle. Shoulder length wavy blonde hair that falls in front of both ears. Blue eyes. A cheerful expression. Kind of tomboyish. Wearing a purple T-shirt and bluejeans. "
I changed the art style from photorealistic to toony to manga to Warner Brothers. I told it "NO spots on the face!"
Here's the results: https://www.crosstimecafe.com/viewt.....795743#p795743
None of them look like my "Kathy." Most have brown muzzles. All of them have spots on the face. When I changed the prompt to include "Tomboyish," it knocked her down three cup sizes.