That’s “Trinitron” 
Figured, but was not sure. Though you can presumably feed it your own input image, (any kind of) noise or not. Not sure if it keeps the same number of total steps in that case.
I tried turning the Perlin noise on and off (on the standalone notebook) but I have not been able to figure out what the ultimate difference is, if any.
That looks to me like a movie poster.
Threw Jeremy Cliff at it and finally got a reasonable approximation of a wheeled vehicle.
Tossed Chip Foose at it and got, well…
Inspired by a thread in FQ, I took a whack at the Kowloon Walled City.
Okay, several whacks.
The idea reminds me of The Rocket Book.
I’m a fan of abstraction, so this one I found quite evocative:
Dancing women, AR Penck, black and white, line drawing
It’s not really quite in Penck’s style, but I do like the use of space and the impression of women dancing.
The yokai portrait tests can be so variable, even between two runs with the same prompt and even the same seed, that often it can be difficult to be sure of a distinctive style without several runs with it.
“Psychedelic art” definitely works. (So do “blacklight”, “day-glo”, and “dope art”.) (eta these are all just psychedelic, it looks like I was saying one of each.)
“Tijuna bible” doesn’t give you simple line art on cheap paper, but it definitely gives you something that counts as a style. I suspect just using “Tijuna” alone would give the same results.
“Andy Warhol” does something. “Chick tract”…might? Again no line-art but maybe the color covers. There seems to be a hellfirey/demony/judgey feel about them. (And I probably don’t need to point out which are Worhol and which are Chick.)
Toulouse-Lautrec and Mark Ryden make an “inseresting” pairing.
(There is some cherry-picking with some of these images)
The death of Pac-Man, by Goya with a few modifiers:
That’s interesting. I posted earlier in the thread about my poor success with Pac-Man.
Amusingly, “Zombie | Style of Gibson Girl” was recognized
Currently obsessing over one image that was thrown out with a “yokai family” prompt—I saw that in an early iteration the “child” looked very much like a dog.
(Earlier iterations often look better than the final, except blurry. I’ve tried this AI on them, but it fails to meet the lofty claims.)
So not only did I do a couple of runs “man with dog” using an early iteration of the image along with the otherwise identical prompt, but I’ve also done duplications, evolutions, start image editing, and varying the prompt to “yokai man with child” and “man with child”, trying to get something with better faces. (In one of them, the child ends up with a very full beard.)
If you want to try playing with it, the prompt is
xxxx | tronie | full-length portrait | hyperdetailed colored pencil drawing | photorealistic | style of dan witz | style of Toulouse-Lautrec
(Where “xxxx” is “yokai family”, “man with dog” man with child, “woman with dog” or whatever) with a seed of 81318, 9:16 aspect ratio and the coherent renderer
(Using the same seed sometimes but doesn’t always throw out a similar image, so it helps to use a iteration 50 as a start image.)
BTW, the site that has this and the super-resolution AI also has lots more programs to play with:
Maybe this AI
will be of some use to you? It’s not one I have tried, though I have used a different, diffusion-based image inpainter to remove unwanted objects from photos (erasing people from the background, etc.)
It works well with Witz.
That’s actually the 75% image as the final result took some unfortunate turns.
These are yokai examples I made. (I didn’t claim that they make realistic humans—we both know that is unlikely.)
Yokai | tronie | full-length portrait | hyperdetailed colored pencil drawing | style of gibson girl | style of dan witz
Yeah, I was using zombie to provide some cover for its limitations. I optimistically tried a couple “beautiful woman” variations afterward and effectively set my credits on fire accordingly.
I got pretty lucky and got a proportioned girl face without a specific artist modifier.
I tried a few combinations to try to get a Cyberpunk Elf Gibson Girl but each time was rewarded with what looks like a cotton candy of face parts
