Artwork Messing around with AI, A Art thread.
- Thread starter Thegoldenboy2188
- Start date
If you want a challenging one, try Lum.Best of four tries at drawing Shampoo in a photorealistic style. I'm on a kick of trying to make old 80s anime and cartoon people but changing their styles to see if I can do it.
Lum (Urusei Yatsura) - Wikipedia
She has some inhuman features (like the horns), so it'd be interesting to see how the AI would tackle those, if at all.
TheRejectionist
TheRejectionist
Bear Ribs
Well-known member
As you wish.If you want a challenging one, try Lum.
There's the original version of her from the 80's/90's, and the Remake/updated design of her from the reboot.Lum (Urusei Yatsura) - Wikipedia
en.wikipedia.org
She has some inhuman features (like the horns), so it'd be interesting to see how the AI would tackle those, if at all.
Horns are a bit difficult if you're going photorealistic due to the lack of real people with horns. This was the best of six inpaints out of six attempts at drawing her, so best of twelve total. If I'd gone for a different style, like Boris Vallejo or a more typical anime, I wouldn't need as many because there's lots of horns in those.
Currently, it's wings and tails that AI struggles with the most, they're the "hands" of the current generation of AI drawing software so the biggest challenge would probably be Morrigan Aesland as the wings coming out of her head are fairly unique and would confuse the heck out of the AI, especially since she also has wings low on the small of her back which is also fairly unique. The combination of four wings all in non-standard locations gives AI conniptions as there just isn't a good base of images to compare with her.
Okay, I have to admit, that's a lot better than what I was expecting.As you wish.
Horns are a bit difficult if you're going photorealistic due to the lack of real people with horns. This was the best of six inpaints out of six attempts at drawing her, so best of twelve total. If I'd gone for a different style, like Boris Vallejo or a more typical anime, I wouldn't need as many because there's lots of horns in those.
Currently, it's wings and tails that AI struggles with the most, they're the "hands" of the current generation of AI drawing software so the biggest challenge would probably be Morrigan Aesland as the wings coming out of her head are fairly unique and would confuse the heck out of the AI, especially since she also has wings low on the small of her back which is also fairly unique. The combination of four wings all in non-standard locations gives AI conniptions as there just isn't a good base of images to compare with her.
As you wish.
Horns are a bit difficult if you're going photorealistic due to the lack of real people with horns. This was the best of six inpaints out of six attempts at drawing her, so best of twelve total. If I'd gone for a different style, like Boris Vallejo or a more typical anime, I wouldn't need as many because there's lots of horns in those.
Currently, it's wings and tails that AI struggles with the most, they're the "hands" of the current generation of AI drawing software so the biggest challenge would probably be Morrigan Aesland as the wings coming out of her head are fairly unique and would confuse the heck out of the AI, especially since she also has wings low on the small of her back which is also fairly unique. The combination of four wings all in non-standard locations gives AI conniptions as there just isn't a good base of images to compare with her.
Would AI assume horns are a sort of decorative headpiece, like a hair ribbon or bow?
Bear Ribs
Well-known member
Depends on the set you trained with. This specific set? Yes.
F'rex suppose you train your AI on the aforementioned Boris Vallejo. If you draw a guy, he's probably going to be huge and muscled by default. If you give him horns, the AI will probably put a horned helmet on him rather than giving him horns on his skull because that's where most of the horns in his art come from. Even if you don't ask for a helmet, in the AI's mind horns and helmets go together on men.
Train on World of Warcraft art instead, though, and it will have a solid understanding of guys with horns growing from their skulls, and will probably avoid putting helmets on them unless specifically told to because relatively few WoW men wear them and even fewer with horns. It will conclude that horns and helmets are essentially contradictory and it's one or the other.
That's why this specific image was a touch tricky and needed twelve renditions, since it was trained on real people it had very few models with horns in the first place and they were all wearing horned headbands so it kept wanting to add a headband. I had to essentially brute force it by putting a very high priority on horns in the positive prompt, and also putting a very high priority to avoid headbands in the negative prompt to get it to understand what was needed. I also had to put a high priority on avoiding cat ears because cat-eared headbands look a lot like horned headbands at a glance, and, I suspect, are far more common than horns, so a solid half of my first rendition actually had the horns as cat ears.
This was a washout from the first set, as an example. You can see the AI was thinking "A headband goes there" but also had instructions not to put in a headband, so it recolored a strip of her hair black where a headband would go.
This incidentally was my second choice. I actually like the face better but didn't want to do a third set to fix the hair overlapping one of the horns so it just barely didn't make the cut. It also has the AI dithering on the subject of headbands and coloring a strip of hair where a headband would go.
Wonder how AI would handle fantasy creatures or really alien beings? I don't mean "Star Trek" aliens which are just variations of humans (which is how they can somehow interbreed, although an old "Star Trek TOS" episode implied we all came from an ancient race), but ALIEN aliens, or even anthro-creatures?
Bear Ribs
Well-known member
ConceptLab actually can do that, it generates entirely new concepts and art styles that don't exist yet rather than riffing on existing datasets the way things like my civilian-grade AI does.
One of my alien-Astral creatures looks like a tall slender furry dragon-vixen with two arms which each split into two forearms, head ruffles on the sides of her head and long red-gold hair. She has a quasi-Asian look about her. Oh, and she has two extra eyes, usually closed, one above the other, on her forehead. Her overall look is red and gold.
Type THAT in to AI and what would it come up with? Or would it blow a fuse?
TheRejectionist
TheRejectionist
As you wish.
Horns are a bit difficult if you're going photorealistic due to the lack of real people with horns. This was the best of six inpaints out of six attempts at drawing her, so best of twelve total. If I'd gone for a different style, like Boris Vallejo or a more typical anime, I wouldn't need as many because there's lots of horns in those.
Currently, it's wings and tails that AI struggles with the most, they're the "hands" of the current generation of AI drawing software so the biggest challenge would probably be Morrigan Aesland as the wings coming out of her head are fairly unique and would confuse the heck out of the AI, especially since she also has wings low on the small of her back which is also fairly unique. The combination of four wings all in non-standard locations gives AI conniptions as there just isn't a good base of images to compare with her.
The only thing I can say is this randomly generated lass looks Korean rather than Japanese.
Speaking of Lum the Invader, this is the British/BBC dub of it of the original series in the 80's.
I'm fucking dying. It's like the Ghost Stories dub.
I'm fucking dying. It's like the Ghost Stories dub.
Man, Bing/DALL-E 3 is better at keeping subjects separate than MJ or SD by far.
Okay, this is one GIANT Sonic, goddamn:
Okay, this is one GIANT Sonic, goddamn:
TheRejectionist
TheRejectionist
Man, Bing/DALL-E 3 is friggin' awesome.
"twilight sparkle standing in a field, with detailed fur, painted in the style of ilya repin, but with cartoonish proportions intact, with a cutie mark of a six-pointed star"
The same prompt, but with Van Gogh:
Rembrandt (this one's missing the horn, but she's just too cute):
Da Vinci:
With ink, in the style of Frank Miller:
Dammit, that's very nice, but that's not what I think of when I think of Frank Miller. Let's try that again.
"twilight sparkle standing on a street corner, in the rain, with detailed fur, done with ink in a dark and moody noir frank miller style, with cartoonish proportions intact, with a cutie mark of a six-pointed star"
No, no, still way too happy.
"twilight sparkle cowering on a street corner, in the rain, with detailed fur, in the middle of a fight with the mob, done with ink in a dark and moody noir frank miller style, with cartoonish proportions intact, with a cutie mark of a six-pointed star"
There we go, finally, goddamn.
Wait, wait, wait, let me do the typography part in Photoshop.
Jesus, poor Twiggies.
I think I'm gonna need a much longer prompt than that. Bing doesn't like short ones. It puts out terrible results.
"twilight sparkle standing in a field, with detailed fur, painted in the style of ilya repin, but with cartoonish proportions intact, with a cutie mark of a six-pointed star"
The same prompt, but with Van Gogh:
Rembrandt (this one's missing the horn, but she's just too cute):
Da Vinci:
With ink, in the style of Frank Miller:
Dammit, that's very nice, but that's not what I think of when I think of Frank Miller. Let's try that again.
"twilight sparkle standing on a street corner, in the rain, with detailed fur, done with ink in a dark and moody noir frank miller style, with cartoonish proportions intact, with a cutie mark of a six-pointed star"
No, no, still way too happy.
"twilight sparkle cowering on a street corner, in the rain, with detailed fur, in the middle of a fight with the mob, done with ink in a dark and moody noir frank miller style, with cartoonish proportions intact, with a cutie mark of a six-pointed star"
There we go, finally, goddamn.
Wait, wait, wait, let me do the typography part in Photoshop.
Jesus, poor Twiggies.
I think I'm gonna need a much longer prompt than that. Bing doesn't like short ones. It puts out terrible results.
Here's a horrifying prompt: "Homeworld Cataclysm/Emergence Beast Infection", maybe with a starship of your choice e.g. "Babylon 5 Fury" or "Covenant Ship, Halo".
Or, ya know, "Doomslayer in Halo": Those poor Brutes!
Or, ya know, "Doomslayer in Halo": Those poor Brutes!
TheRejectionist
TheRejectionist
Communist Japan street with civlians and soldiers on a sunny day @*THASF*
Bing has a very overactive filter that blocks real-world figures, political content, et cetera. No matter how I vary the prompt, I get their stupid warning about how the output was blocked.
So, I used Midjourney instead, which had pretty good results.
"Poster of Japan city street with soldiers and civilians on a bright, sunny day, done in the style of soviet posters, in a hypothetical alternate universe with a happy and prosperous communist Japan --ar 9:16"
TheRejectionist
TheRejectionist
Holy shit thàt is impressive, especiàlly in the left, I càn tell from the humàn figures the got them probàbly from DPRK, CCP & Communist Southeast Asia countries. I sày this becàuse on the right imàge you càn see more Koreàn looking àlphàbet.Bing has a very overactive filter that blocks real-world figures, political content, et cetera. No matter how I vary the prompt, I get their stupid warning about how the output was blocked.
So, I used Midjourney instead, which had pretty good results.
"Poster of Japan city street with soldiers and civilians on a bright, sunny day, done in the style of soviet posters, in a hypothetical alternate universe with a happy and prosperous communist Japan --ar 9:16"