That link goes to your picture, but it looks like this is your new thread:
I’ll have to see if I can get something like that running locally. Occasionally I get images like this:
The right half is fantastic, and exactly what I wanted. The left half is… I dunno what happened there. But I’d definitely like to crop it off and regenerate that part.
Woops! Thanks for the fix.
Some times in stable diffusion it tries to render two angles of the same subject in one frame, which is interesting. It is almost like it has an idea of objects in 3D.
One article about Dalle-2 had some excellent images generated with “A hybrid between Spongebob SquarePants and Godzilla, painted by William-Adolphe Bouguereau”. So far I haven’t quite acheived greatness in stable diffusion. (Spoilered for NSFW content)
One “zany” prompt I’ve tried in different sytems is “Alf and Yoda eating lobster”. Both Craiyon and Min-Dalle clearly know who Alf is and what he looks like (even though they do a pretty terrible job of drawing him).
Stable Diffusion, on the other hand, is very confused. It tends to render (when I renders “him” at all) some sort of brown anthropomorphic dog or Furry, even when I specify “Alf from the TV series Alf (1986)”. The attempts tend to be incredibly awesome, though.
I’m thinking Stable would be very good for creating Vargas Girls. When it doesn’t add extra parts, that is.
Once upon a time we were trying to get Nightcafe to draw penanggalan. I can’t find the post because there’s too many ways to spell penanggalan but I did get some pretty rockin’ pictures out of MidJourney. I had to describe what it was though as it also wasn’t familiar with the word.
Spoiler tagged just because it IS a floating head with gross stuff
https://mj-gallery.com/dfdb8658-1857-4b38-b056-29ef2fb8206e/grid_0.png
https://mj-gallery.com/3efc4c41-fe08-4b2c-8d0e-ff4338801ca1/grid_0.png
It seems they have update the Stable algorithm to include a start image. I’ve been messing with it and I’ve come up with a few decent results, but nothing really predictable yet. I’m not sure I really get how it fits in to the final product. Very hit or miss for me so far.
It was me attempting, using “karsue”. I tried it in stqlable diffusion, to total failure.
Months ago in “coherent” I tried a prompt for a Muppet family portrait. The results were mostly mangled. I recently tried the same prompt in “stable” and the output was much more interesting, with a mix of decently made Muppets and people (including John Travolta and a giant-eared Dan Rather). Then I tried for puppet family portraits. Those turned out to be really freaky, generating weird aliens and creatures that look pretty much nothing like puppets. (I also tried for awkward family portraits, but—while definitely awkward—they don’t rise to the level of awesomeness from the puppets.)
These are a cherry-picked mix of results from all three prompts
x family portrait | tronie | hyperdetailed colored pencil drawing | Dan Witz | Margaret Keane | Zdzisław Beksiński | Dungeons and Dragons guidebook | postcard
Once again, thanks for sharing some intriguing efforts. Will have to try it.
Hearing that I had to give it a try. Used a starter image that was generated also using Stable and using the same prompt, the results were similar, but just different enough that I want to try multiple generations of it. Will need to wait for more points to build up.
https://creator.nightcafe.studio/creation/AWmBkMb4oOu15TDQvR3n
I just tried to see if it would extend/repair images in the method shown in that Dalle-2 manual. It does not. Completely unrelated additions.
Main image is safe but there is at least one in the group that isn’t exactly SFW.
Are you talking about this?
Was there something you did to get the Marilyn-Monroe-Seven-Year-Itch effect, or was that just the AI’s interpretation of what goddesses look like? And what are the Asian characters in the prompt?
I mean like in post 1080, but manually using the method used in that Dall-E guide PDF. I wanted to extend badly framed images with matching content, but it failed spectacularly.
I am not a comics expert, but there is a Wiki article describing “少女漫画” as girls’ comics, with distinctive visual styles influenced by illustrations accompanying “flowery and emotional” novels for girls and sentimental lyrical painting influenced by Art Nouveau and Japanese-style paintings. So you are not going to get classical representations of Artemis.
Yes, the Japanese is “Shojo Manga” or girls comics. “Clamp” is also a famous girl’s comics artist (actually a group of artists) that I thought might be famous enough to be in their training data. Other times I’ve tried a Japanese prompt I got vaguely oriental nonsense, so I think the training data is mostly in English. I’ve liked what this prompt has given me, it does seem to be in the right style.
“young goddess in a dress of leaves” is inspired by an illustration from a children’s book I remember from long ago. Don’t remember anything else about that book, just that one image. Might have been a fairy instead of goddess.
The seven year itch effect is something it came up with on it’s own. Others from the same prompt were more modestly dressed, but I liked how that one turned out best.
