Stable Diffusion, NovelAI, Machine Learning Art - AI art generation discussion and image dump

This is actually happening because the AI was trained on 512x512 resolution pictures and different formats can but don't necessarily work well as the AI basically "forgets" what it was doing midway through the picture. The current webui has a workaround for this.
It is funny: all of the “cursed” clone booba photos getting made into memes are basically coming from users making the AI go outside normal operating range (intentionally or not). This is good to know as we can force the AI to make cursed images for teh lolz (or avoid it altogether). Lots of room to play here.
 
Thought I'd do a quick generation of one of our lolcow harrassers, Liz Fong-Jones:
1666141630344.png
 
The most concerning thing that I've seen people have a reasonable fear of isn't of being generally replaced- even if an AI could do literally anything, it wouldn't replace artists, in my opinion. The more tangible fear is of artists being replaced by themselves. There's already been several artists who have had their work fed into algorithms to produce material similar to what they make- and utilizing their name as a tag to specifically get it. Fascinating as that is, it's pretty threatening to be at risk of being run out of your work by a mechanical duplicate of yourself, and angers people too, because their work was put through a machine to duplicate it without their consent or knowledge.

Solving that issue of random people using an artist's own work to nullify them isn't something to just snicker at I think. The one thing I think could be hugely damaging to the algorithm development scene is people getting sick of their shit getting ripped and fed to learning machines, so they take it to court and put the development company through the wringer for something along the lines of using stolen intellectual property. After all, the algorithms have internal references, and if they lose a huge chunk of them, then you're back to square one. I think that's something that AI developers have to get ready for the possibility of.
I can't lie that I'm getting a giddy off of NSFW artists who paywall their stuff behind Patreon Discords. Too many of them moan on social media about how hard their lives are when they're making multiple grands a month on top of commissions, then they post 2 sketches in a month and want you to pay 10 dollars a month for that.

It's not something people are going to avoid either, even if you paywall your work, sites like kemono will always exist to rip it because people hate paywalls.
 
Specifically this tech doesn't bother me, but rather the potential for what could possibly be next
 
This is actually happening because the AI was trained on 512x512 resolution pictures and different formats can but don't necessarily work well as the AI basically "forgets" what it was doing midway through the picture. The current webui has a workaround for this.
1A1B7C86-FFCF-4D87-8023-99B9F3567D55.jpeg

Mi brudda, now be a good time to make peace with Jah. Dark times be ahead.
 
Somebody generate me a fun image of Zyklon Gift Gas. It can be anthropomorphic or whatever, I just need something for my name sake.
 
  • Like
Reactions: Tree
View attachment 3748761

Made a waifu Dante. I'm scared.
Out of curiosity, I tried making an accompanying guide for him. While just using "Virgil" seems to generate just a bunch of anime girls, putting "Roman poet Virgil" primes the AI hard into statues and busts. I mean, the other kind of busts. A bit different than the one outside his crypt, but not bad.
Roman poet Virgil s-66114971.pngRoman poet Virgil s-793619476.pngRoman poet Virgil s-2352027550.pngRoman poet Virgil s-2603422651.png

Putting in "Dante's Inferno", other than generating a bunch of anime tits and beefcakes, also generated some interesting bits. Lots of views of massive architecture seen from below along with fire, naturally. You can see the model's preference for humanoid forms coming through.
Dante's Inferno, s-56947283.pngDante's Inferno, s-2138838128.pngDante's Inferno, s-2496712887.pngDante's Inferno, s-4034960124.png

These are only single tags and letting the AI do its thing, but with some more precise prompting and the use of in-painting, I could see these looking fairly nice. I'll be interested in testing and seeing if the AI can manage some artwork in the style of Zdzisław Beksiński, with or without the NAI anime-tuning.
 
Back