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

[Starved Artist]

View attachment 4301705
Don't know if this has been posted yet, but this editorial from 1859 certainly sounded familiar.

>Painter crying about photography replacing them.

Failed Artist Complaining about Ai Art.

History repeats itself far too often for my likes. We're supposed to breaking new ground here, but every other year, we just relive bullshit people from a hundred years ago handled far better than we are.

 

[Epic Fail Man]

These niggas complaining about A.I. know nothing about art. Part of going into a gallery and looking at paintings or projects is to look at the details and find the intent or meaning of each characteristic of their creation and trying to interpret what kind of emotion or message they were putting into their work.

A.I. has no intent or purpose on any detail it generates, it's random. it is the equivalent of watching two crackheads fight each other rather than look at a message created in an abstract form, that's why it's so fun because you don't know what it's going to produce.

If you think it's anywhere close to replacing art or has the same appeal you are a fool and should probably go to art school and learn why people draw and paint. Anyway here's Terry Crews hanging out with Pikachu.

 

[Pendulous Dong]

Goddamn this is fun. Once you get a handle on it, you have the ability to really fine-tune this stuff to be exactly what you want it to be.
Here are a couple of pirates.

 

[Still Exceptional]

I still haven't figured out how to set up Stable Diffusion on my PC, but I've gotten good results from some of the free sites out there. I'll post some of my best results.

Spoiler: Big post

Lovecraft style "Eldritch God"


Death Knight


Undead Thief



Undead Lord



Female Berserker


Demon Lord


Bearded barbarian berserker in armor

Honestly, this shit would be absolutely great for the artistically disinclined tabletop player.

 

[purpleflurp]

SD on its own doesn’t seem to have a lot of references to alien terrain in the original dataset, so I built a quick Dreambooth model to try and add some more variety. Dataset was 21 screenshots from Black Mesa’s Xen trailer, with 300 classification images.

Train 80 epochs per image at 5e-6 with a polynomial learning rate scheduler and Dataset size *8 warmup steps.



Not my best work, but it’s a start

 

[magnetic field]

amen.

 

[The Mass Shooter Ron Soye]

I have a request: glowing glass chalice full of milk being held by Alex Mahan

Spoiler: something like this please...

View attachment 4308356View attachment 4308359

Begging thread here: https://kiwifarms.net/threads/ai-image-requests.130976/

 

[AmpleApricots]

This juggling of different models/trying them out/maybe sometimes combining them is like skyrim modding all over again.

 

[Monkey Pink]

Is there some way I can give an AI a picture of a character, and it would generate prompted pictures of the character doing X or Y? Like for fan art? Not quite img2img, is the technology there yet?

 

[The Mass Shooter Ron Soye]

Is there some way I can give an AI a picture of a character, and it would generate prompted pictures of the character doing X or Y? Like for fan art? Not quite img2img, is the technology there yet?

That sounds like a job for DreamBooth.

 

[Monkey Pink]

That sounds like a job for DreamBooth.

I'll give it my best shot.

 

[purpleflurp]

Is there some way I can give an AI a picture of a character, and it would generate prompted pictures of the character doing X or Y? Like for fan art? Not quite img2img, is the technology there yet?

Like Mass Shooter said, you need a dreambooth model to do this.

Depending on the style you want to do, I think 20-30 images at 1600 training steps will suit your needs. Original source material works best, but you can make do with high-quality fanart.

 

[Muu]

 

[Still Exceptional]

The Virgin Human Artist:
Ugly, bland
No talent required, a kindergartener could do it
Does not provoke any emotions but contempt for the creator
Literally a scam to launder money
Is somehow considered art by snobs


The Chad AI Artist:
Beautiful, vibrant
Height of talent, you probably couldn't do better
Provokes a plethora of emotions
Made because someone wanted it to exist
Is somehow not considered art by snobs

 

[Abeezo]

I dont really post nsfw very much but I am making an exception.

Spoiler: Warning lewd stuff and boobies

Also I'm curious if you guys have created any OC's from ur outputs? I was talking to a buddy of mine if people did that with AI art

 

[Mister Qwerty]

It has been foretold…

 

[The Mass Shooter Ron Soye]

It has been foretold…

View attachment 4314096

I glanced at this on Ars Technica but I skimmed over the fact that it was predicting the first full year of the AI art mania:

https://arstechnica.com/information-technology/2023/01/a-cartoonist-predicted-2023s-ai-drawing-machines-in-1923/

https://archive.ph/CcKkk

 

[purpleflurp]

Also I'm curious if you guys have created any OC's from ur outputs? I was talking to a buddy of mine if people did that with AI art

I’m still reading a few papers to get this figured out.

Right now I can replicate a character present in my dataset through tagging, but I’m assuming creating a new character in the AI will eventually require retraining the model with those generated images.

Loathe as I am to credit lolicon fans, the people behind the Made in Abyss model are probably on the right track to consistently generating detailed characters.

 

[AmpleApricots]

I was out a bit of the whole thing for a bit and started getting into the newest stuff again since a few days here and there and the development jump from just a few months ago is kinda crazy tbh. Feels kind of like computers some 30 years ago where there was no telling what things would be available in two years.

Also coomers man, I think they drive most of this. I've seen convincing nudes on civitai and reddit where if shown in a blind test, I would not be able to tell you if these are real photos (albeit photoshopped to airbrush etc.) or AI generated. This was not a thing a few months ago, nor was it possible. It's a completely unregulated frontier but somehow I start to think it won't stay that way for long. I'm convinced we're at a point where you could for example generate photo-realistic child porn that'd stand up to some scrutiny. (Frankly, I'd be surprised if people don't already make such) Yet it is generated. What does that mean legally? These are things that are gonna be decided in courts in the next years and I wonder what the impact will be. (Also very apropos to the early computer and internet times where "the law" basically didn't understand that stuff and you could get away with *a lot*)

 

[The Mass Shooter Ron Soye]

I'm convinced we're at a point where you could for example generate photo-realistic child porn that'd stand up to some scrutiny. (Frankly, I'd be surprised if people don't already make such) Yet it is generated. What does that mean legally? These are things that are gonna be decided in courts in the next years and I wonder what the impact will be. (Also very apropos to the early computer and internet times where "the law" basically didn't understand that stuff and you could get away with *a lot*)

It's already forbidden by the PROTECT Act's "indistinguishable from that of a minor" language. Somebody might try to weasel out of charges for possession of the image model.

The real battle could be over deepfake porn of adults. The revenge porn laws probably don't apply, and the deepfake laws only appeared around 2019. Kamala Harris has gone after revenge porn websites in the past, so she might want to get involved.

SCOTUS hasn't touched on right of publicity laws. If those laws were struck down, maybe Disney could deepfake actors without paying them or their estates.