https://app.wombo.art
I saw this thing used on Reddit to create a few odd but also accurate movie posters for existing movies, so I decided to introduce it to B12 as well.
It's rather creative, usually accurate and often makes beautiful pictures, although it has... issues... with drawing faces and bodies up close.
I won't lie, I'm kinda interested what kind of monstrocities Bay 12's collective efforts will be able to create.
---
I put a few of the pictures that I generated with it here (https://imgur.com/a/uKqsBZf).
Its allegories of various countries turned out to be hilariously stereotypical while still very weird, so I made a separate imgur post for them (https://imgur.com/a/4IcUDuw).
Four wonderful examples from the two above links (warning: large images):
(https://i.imgur.com/JM4CjeH.jpeg)
(https://i.imgur.com/RF2FBj5.jpeg)
(https://i.imgur.com/E1E8q5P.jpeg)
(https://i.imgur.com/mhcKIU5.jpeg)
And, uh, it seems to have read a very different version of Alice in Wonderland than most of us did:
(https://i.imgur.com/XL0gYWV.jpeg)
(https://i.imgur.com/XCaibJo.jpeg)
(https://i.imgur.com/Z0eUGq7.jpeg)
Is that what Alice in Wonderland looks line when you take some acid before reading it?
Perhaps, if it was a really bad trip.
Or maybe that's what Lewis Carroll would have written if he had access to LSD instead of opiates.
Here are some more examples (https://imgur.com/a/pVMroyE):
(https://i.imgur.com/adU3Ap3.jpeg)
(https://i.imgur.com/MCeCdtK.jpeg)
(These remind me of the Hugs and Kisses Elemental, personally. In the second one, you can see all those teeth barely fitting under a thin veneer of cat.)
(https://i.imgur.com/TE4Pgp5.jpeg)
(https://i.imgur.com/ERxzSPQ.jpeg)
(https://i.imgur.com/XjI3lYm.jpg)
Let's try with one of the favorite people of the bay12 posters from the american politics thread :
(https://i.imgur.com/B22CrBq.jpg)
As it turns out, the AI works pretty well with long phrases. Such as quotes.
It appears to be using pretty advanced language processing - not nearly as advanced as DALL-E, but it separates concepts out of a phrase in a shockingly precise manner.
Prompts can be seen at the bottom of each picture.
https://imgur.com/a/gD5wav0
(https://i.imgur.com/p1rfdlz.jpg)
(https://i.imgur.com/C12fcRq.jpg)
(https://i.imgur.com/UCOJvyY.jpg)
The original quote (slightly shortened to fit 200 characters) is far less spectacular, probably because "Durin's Bane" is not as obviously linked to Balrog-like images:
https://imgur.com/a/KNGSUAE
(https://i.imgur.com/dz205rD.jpg)
(https://i.imgur.com/e1viUWV.jpg)
And it can imitate styles, too. But it does not parse the sentence's structure the way DALL-E does, unfortunately.
The results of my tests are here (https://imgur.com/a/q2SvIsA). I'm afraid they're a bit too disturbing, even compared to the Cheshire Cat, so no spoiler tags for them.
UPD: It doesn't merely search the Internet for pictures matching the keyphrase - it somehow fishes out the most meaningful parts.
Otherwise it wouldn't have generated this:
(https://preview.redd.it/i7if6h1h6q281.jpg?width=1080&format=pjpg&auto=webp&s=f426e90e2c958e8c1d4f1a697037f26bff817b01)
My first guess was that it's trained on separate concepts (potato, jesus, potato jesus) and prefers ones that are less complex to more complex ones.
But then it would parse "Twilight" and "Sparkle" as separate concepts when I put "Twilight Sparkle" into the prompt - instead, it painted a pony of the same name.
So, my first guess was wrong. There has to be some kind of language-processing technosorcery going on.
(https://cdn.discordapp.com/attachments/878129313722015754/915845446017105920/Fluffy_Wambler_TradingCard.jpg)
Ancient problems require modern solutions. (https://imgur.com/a/DvzddWv)
(https://i.imgur.com/omp9O05.jpg)
(https://i.imgur.com/zmE2cnT.jpg)