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

As for me, I haven’t made much other than some projects where I try to learn different plugins and just updating the archive. Sorting the archive of models and loras is such a huge pain in the ass, but I want to share my favorites at some point, with examples of what they can do, but this will take a while. I hope you are all having fun and I hope you guys will share more of what you guys are creating or working on. Much love.
I'm actually setting up a Huggingface account that I'll be uploading my models to. I've gotten written permission from several of the artists I've been building models from for personal training and use, but there's a few I have on deck that'll be fine with me posting public models for.

I'd be more than happy to link you to the models I upload so you can add them to your archive. I doubt my work's enough to attract vultures trying to scalp models for their apps, but you never know what'll happen to these sites.

I really love looking at the progress you are making. It’s really fun to see over these posts how the pictures change as you move closer to the goal. Very good job!
The big thing driving me with these experiments is I really fucking hate how most of the "creators" using AI art are just trying to play it safe playing using NAI/Midjourney styles. But due to how the training for these models works, the actual "soul" of the art gets smooshed together and homogenized as different vibes and aesthetics get overwritten and averaged out.

Like this looks cool
stormtrooper.jpg
But to me, as someone that likes collecting giclee prints, I don't feel anything when I look at it. It's a cool image, but it feels like looking at a graphic tee at Target: Neat design, but it won't stick with your mind.

Compare that to something like this:
volrath.jpg
This is unsettling, and when I bought the print of it a few years back I felt kinda gross handling it. It's probably a result of me being a weirdo, but when I pick up the print, I get like this uncomfortable feeling when I actually touch the art. Now, in my opinion there's a reason I feel this way beyond being a fucking weirdo:

Anson Maddocks, the artist of this piece, built his style off of surreal anatomical horror a la HR Giger. When he paints a character, it's very seldom a "beautiful" character: His elves have features more fitting for vampires and ghouls, his minotaurs have intricate warpaint and horn carvings, and his humans are often gaunt and pale. Even in cases where the character is "beautiful", the painting will often be cold and melancholy or sickening and disturbing.

When you look at an Anson Maddocks painting, you can immediately recognize it as an Anson Maddocks painting because of the style that he employs with his works. And when you look at the painting, whether it's on a card or print, you can appreciate the unsettling vibes it gives off. Even with abstract art like Maze of Ith, you can see the same concepts in it that you see in the Volrath art I posted above.
5107-maze-of-ith.jpg

It probably sounds autistically self-righteous, but when I'm building these AI models I'm trying to train and tweak them to create art that can make you feel something when you look at it. Even if I'm using another artist's work as the brick and mortar for a model, I want to insert my own ideas and preferences into it so that the art it creates can be recognized and stick out from the rest of the typical AI art that's churned out from the other models.
12.png5.png

There's also the fact that I myself am a watercolor artist, and I'd argue a pretty decent one. I've got a good grasp on composition and color theory, but there's a few things that I don't like painting or can't do quite right. For instance, I and many others absolutely hate drawing complex mechanical things like mechs and clockwork machinery. But if I use an AI model trained on my natural art style to generate the lineart for, say, a Gundam or SCP-914, I can create the basis for a complex scene that I myself couldn't visualize normally. I can then do the lineart I can draw normally, transfer the AI lineart onto the watercolor paper, and continue on with my normal inkwash.

And most importantly, I think the juxtaposition of being a user on a site like Kiwifarms going between laughing at trannies/lolcows/etc and working passionately to share their artistic endeavors while helping others do the same is absolutely hilarious. Doubly so when I put more effort into perfecting my work than most artists sneeding about AI have
 
Last edited:
Thank god traditional animation never has weird artifacts
View attachment 5044738
View attachment 5044744View attachment 5044747View attachment 5044750

Just render the girl naked and animate the clothes on a second layer, redrawing when they aren’t good enough. It’s literally what I’d do when animating a character wearing stuff that needs simulated physics
That's not artifacting; those are drawn intentionally, and known as "animation smears." They're put in by the animators intentionally to visually smooth out movements, since it would look choppy otherwise. When played as part of the animation it's much less noticeable because it's only a frame.
 
That's not artifacting; those are drawn intentionally, and known as "animation smears." They're put in by the animators intentionally to visually smooth out movements, since it would look choppy otherwise. When played as part of the animation it's much less noticeable because it's only a frame.
It was meant to be a tongue in cheek response to how artists were cherrypicking individual frames to write off the tech, but I realize I could've used a better example.
 
I'd be more than happy to link you to the models I upload so you can add them to your archive. I doubt my work's enough to attract vultures trying to scalp models for their apps, but you never know what'll happen to these sites.
I'd appreciate that. I'll def add it to the archive!


The big thing driving me with these experiments is I really fucking hate how most of the "creators" using AI art are just trying to play it safe playing using NAI/Midjourney styles. But due to how the training for these models works, the actual "soul" of the art gets smooshed together and homogenized as different vibes and aesthetics get overwritten and averaged out.
This is unsettling, and when I bought the print of it a few years back I felt kinda gross handling it. It's probably a result of me being a weirdo, but when I pick up the print, I get like this uncomfortable feeling when I actually touch the art. Now, in my opinion there's a reason I feel this way beyond being a fucking weirdo:

Anson Maddocks, the artist of this piece, built his style off of surreal anatomical horror a la HR Giger. When he paints a character, it's very seldom a "beautiful" character: His elves have features more fitting for vampires and ghouls, his minotaurs have intricate warpaint and horn carvings, and his humans are often gaunt and pale. Even in cases where the character is "beautiful", the painting will often be cold and melancholy or sickening and disturbing.
It probably sounds autistically self-righteous, but when I'm building these AI models I'm trying to train and tweak them to create art that can make you feel something when you look at it. Even if I'm using another artist's work as the brick and mortar for a model, I want to insert my own ideas and preferences into it so that the art it creates can be recognized and stick out from the rest of the typical AI art that's churned out from the other models.

I feel the same way. Sometimes i like to make normal appealing art with the AI, mostly NSFW or specific shit. But for the most of the time I take loras and put them on top of other models and mix styles that shouldn't be mixed. I started learning how to make AI art just so i could create body horror and other horror art. I've been eternally haunted by Event Horizon and Hellraiser so i wanted to make stuff that resembles that. I'll share some of them, not sure if anyone is interested in this kinda stuff.

It goes without saying, but these are not safe for work.

00080-2527494181.png00082-3948360470.png00083-398929881.png00084-398929882.png00086-3392268604.png00087-2364208785.png00088-2364208786.png00089-2529026851.png

IMG_2778.jpgIMG_2779.jpgIMG_2780.jpgIMG_2782.jpgIMG_2783.jpgIMG_2785.jpgIMG_2794.jpgIMG_2795.jpgIMG_2833.jpgIMG_2839.jpgIMG_2840.jpgIMG_2841.jpg

Serious warning here. Extreme body horror below
IMG_2828.jpgIMG_2829.jpgIMG_2830.jpgIMG_2835.jpgIMG_2836.jpgIMG_2844.jpgIMG_2845.jpgIMG_2846.jpgIMG_2847.jpgIMG_2848.jpgIMG_2875.jpgIMG_2876.jpgIMG_2877.jpgIMG_2880.jpgIMG_2881.jpgIMG_2883.jpgIMG_2885.jpgIMG_2887.jpgIMG_2888.jpgIMG_2890.jpgIMG_2891.jpgIMG_2892.jpgIMG_2893.jpgIMG_2894.jpgIMG_2895.jpgIMG_2896.jpgIMG_2897.jpgIMG_2898.jpgIMG_2899.jpgIMG_2900.jpgIMG_2906.jpgIMG_2907.jpgIMG_2908.jpgIMG_2912.jpgIMG_2914.jpgIMG_2915.jpgIMG_2916.jpgIMG_2917.jpg
 

Attachments

  • IMG_2921.jpg
    IMG_2921.jpg
    453.5 KB · Views: 18
  • IMG_2920.jpg
    IMG_2920.jpg
    843.9 KB · Views: 19
  • IMG_2919.jpg
    IMG_2919.jpg
    845.8 KB · Views: 22
  • IMG_2918.jpg
    IMG_2918.jpg
    1.1 MB · Views: 22
  • IMG_2922.jpg
    IMG_2922.jpg
    1.2 MB · Views: 29
I've been troubleshooting an issue with the image becoming fucked up on the last step, like, really blurry or distorted. So far Hires fix helped with that, I got it to upscale via that Erdogan whatever its named upscaler. Downside is that it really hogs CUDA memory and I gotta do a full PC reboot to make it work again. Anyone dealt with that before?
Share some wisdom with me please, I will give some japanese shitfaced goblins i proompted in exchange.
 

Attachments

  • 1682794022105.png
    1682794022105.png
    1.6 MB · Views: 36
  • 1682794032289.png
    1682794032289.png
    1.6 MB · Views: 36
  • 1682794041469.png
    1682794041469.png
    2 MB · Views: 38
I've been troubleshooting an issue with the image becoming fucked up on the last step, like, really blurry or distorted. So far Hires fix helped with that, I got it to upscale via that Erdogan whatever its named upscaler. Downside is that it really hogs CUDA memory and I gotta do a full PC reboot to make it work again. Anyone dealt with that before?
Share some wisdom with me please, I will give some japanese shitfaced goblins i proompted in exchange.
I've had the issue but i can't for the life of me remember what i did to fix it. I have some suggestions tho if you are down to troubleshoot some more.

1. Antiburn/Burn
There are two issues that might be affecting it and that is "burning". If you have the CFG scale set too high for the model or lora, it can "burn" the image. Usually this shows up as blurred images or too sharp images. The color is often the telltale of it as it will often blow out the color saturation, making the image insanely oversaturated. What is your CFG scale at? Mine is usually at 7-10 depending.

Also if you have the Anti-burn plugin it might be fucking up the image at the last second
1682794753320.png

2. VAE
It could be a fault with the VAE if its not baked into the model. Are you using a specific VAE? Have you tried changing the VAE? I usually have mine on auto, but i change from time to time when i meet issues with either the picture being blurry or lacking contrast/saturation1682794605815.png

3. The sampling method is scuffed, try changing the sampler and see if anything changed
1682794803860.png

4. The model is made for higher or lower sampling steps, try changing it around and see if anything changes for better or worse. I usually keep mine at around 40-80 unless i go for a very 2D style.
1682794851768.png


5. Its a 2.1 model and you don't have the .YAML file that 2.1 models need, which can cause issues

6. Its a ClipSkip issue. Try changing it to 1 if you are on 2 or changing it to 2 if you are on 1. The general ClipSkip rule is:
ClipSkip 1 is for Realistic
ClipSkip 2 is for Anime
1682794967627.png

That is what i have for now, if i remember something more i'll add
 
Last edited:
That was quite handy. I think I narrowed it down to a .vae issue. I didn't have it set to automatic. Here's a Suika without any upscaling.
View attachment 5099792
Hope its fixed :)
Some models have the VAE built into them and will pick those first if its set to Auto. I usually save a screencap or pdf version of the place where i downloaded the model just so i can check there if they recommend any settings or specific VAE before i begin. Other than that i just randomly click buttons until something cool happens.

Here's a nightmare i made while testing RED filters and violence filters on an online service. Probably doesnt need to be said, but their filters weren't that good.
IMG_2792.jpg
 
to be fair, these are smear frames
All the same you can find tons of artifacts that aren't intended in traditional animation, and in digital animation that has long since been switched to. Ai has a lot more problems than most human made animation right now but eventually the problems will equal out.
 
I started learning how to make AI art just so i could create body horror and other horror art. I've been eternally haunted by Event Horizon and Hellraiser so i wanted to make stuff that resembles that. I'll share some of them, not sure if anyone is interested in this kinda stuff.
This is easily some of the most visually interesting and inspiring stuff I've seen in months, AI or otherwise. Well done.
 
  • Feels
Reactions: Psicopax
Some new ones:

Rockabilly biker hyena
rockabilly biker hyena.png

Elvis impersonator sasquatch
Elvis impersonator sasquatch.png

Pirate Yeti
pirate yeti.png

Clown orangutan relaxing on the beach
clown orangutan relaxing on the beach.png

Yeti Ska band
yeti ska band.png

Space pirate crabs
space pirate crabs.png

Carnotaurus eating cotton candy at the circus (my favorite of this batch)
Carnotaurus eating cotton candy at the circus.png

Cyborg samurai penguin
cyborg samurai penguin.pngcyborg samurai penguin 2.png


@Psicopax yeah I just use the standard free Stable Diffusion generator.
 
over on a YT video about a threat to the Internet Archive site, there's this comment:

As an artist, I have to say that copyright laws should stay because of stuff like AI image generators, but copyright laws should not be able to stop people from archiving things.
(of course an artist would say that)

also the video in question:

"What is Happening to The Internet Archive?" - All Things Lost
 
Back