Makes sense though. A “better” 1.5 without the load requirements of SDXL sounds like a winner.
Night Café just gave me some nice Sci-Fi book covers.
Is that with one of the new loras they added?
Just SDXL 1.0.
Using only the word “what” as a prompt
Better AEN’s than on the Mad Magazine cover images posted recently.
Here’s a new toy to try.
Looks like basically a “build a prompt” interface running over stock SD1.5? I’m guessing, anyway, based on the image and face quality. Some of those faces look like they could really benefit from a pass with CodeFormer.
Playground AI just implemented (or at least, I just noticed it) an option called, “Expand Prompt,” which they explain as, “Use AI to improve short prompts and get new image style ideas.”
So basically if you put in a prompt for “Scary Face” you get this (SDXL - Delicate Diffusion):
But if you click, “Expand Prompt,” you get this (same filter and seed):
This is the actual prompt it came up with: “Captivatingly eerie face that has taken ArtStation by storm, shot with meticulous precision in a professional studio setting by renowned artist Greg Rutkowski, leaves nothing to the imagination with its extreme levels of detail, sharp focus, intricate details, digital painting, ultra clear, ultra realistic, breathtaking surreal masterpiece”
It actually adds to the prompt with related(?) words, but you can’t edit it; you get what you get. You can always copy that style in a new image, of course, so that’s pretty cool.
Bing/Dalle 3 is getting good with stylized cartoons. Serviceable with a some minor Photoshopping:
DE3 does not know what the cover of Action Comics #1 looks like.
Context: There is an AI image that went viral lately of a cat stealing a fish that (of course) lots of people think is real.
https://www.reddit.com/r/ChatGPT/comments/17uau9f/how_do_i_generate_something_as_marvellous_as_this/
So I’ve been having fun with variations on the theme today.
A variation of the variation:
One of my favorite recent discoveries is using “humanculus” in prompts for Bing/DE3. You can get some incredibly weird and incredibly NSFW results. I’ve made over 600 of them, here is a large gallery full.
Those are great. I can’t stop laughing at some of those monstrosities.
Aside from the scuba diver having three arms and a disembodied extra fin, and the fish not actually having a cat in its mouth, that’s spot on.
@Darren_Garrison , that’s interesting in that it clearly knows that Action Comics #1 was a Superman comic, and it knows pretty well what the cover of a Superman comic looks like, but it’s not the right Superman cover. IIRC, AC1 had him holding a Beetle-ish car over his head?
(Googles)
Yup.
EDIT: Even more interesting: It got the design of the title basically correct. How the heck did it get the title right without the right image? #1 is by far the most famous issue of Action Comics-- It almost certainly got that title from images of the actual Action Comics #1.
A sample prompt (hidden for reasons).
Note that at no point did I ask for a vagina face and ballsack.
Also notice how it doesn’t mind trying to reproduce US $100 bills.
It is almost Turkey Day…
The first image is poor. The other 3 (scroll right) are a little better.
Created with Bing AI