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

1666121404395.png
Who could this lad be?
1666121476486.png

LOL thanks, the deepart fuckup spurdo is my new avatar.
1666121494622.png
 
This AI art technology makes me think of the transition of warfare from knights on horseback to the era of gunpowder but with art. Now anyone can make something aesthetically pleasing with some key words and a PC. The evolution of this technology is going to be interesting to see.
 
Am I the only one who thinks the generic anime girls this ai generates look uncanny
I agree, it's getting better at it, but surprisingly it's way more 'talented' at producing realistic artwork than cartoonish stuff. If you train it to generate something in the style of a renaissance portrait it will do wonders, if you ask for something that looks like a realistic 3d render, it will get you good results 1/3rd or 1/5th the time at worst. Cartoons end up getting distorted obviously in some way or another, but perhaps it's just it being more obvious to the human eye when something simple has imperfections.
 
No one is making an AI Bach for new Baroque pieces, even though programming-wise it would be trivially easy, as the rule sets for that sort of music are very narrow.
Well, someone did.
From 2017.
 
No one is making an AI Bach for new Baroque pieces, even though programming-wise it would be trivially easy, as the rule sets for that sort of music are very narrow
There was a Bach google doodle a few years back that did exactly this. You would input a few random notes and it would auto-generate chord sequences based on them to create a harpsichord melody
 
Last edited:
what practicality does this have except making surreal shitty album cover looking art.
It is adept at producing repeating patterns and interesting textures. Already I have found a spot for it in my graphics workflow. Oh the time I've spent creating patterns in illustrator, and this thing can replace much of that. The creative possibilities are endless.

Over the next few years this will be refined to be able to produce worthwhile video content, and I think that will be the real game changed with this tech.
 
Being crippled precludes you from walking, not being good at art doesn't preclude you from learning how. Being bad at art doesn't make you crippled.
My mistake, I was not aware that an analogy would be an obscure concept to you. Sorry if it puzzled you.

Next time you want to put me in the deep fryer, I would appreciate it if you'd at least try to comprehend my actual meaning* before you do it. That's all I have to say.

Edit: For the records, I meant "comprehend my entire post", not the insignificant piece you chose to target for the sake of hostility.

Make a basic MSPAINT sketch (brushes add noise so they are better) and you can probably do something with img2img.
View attachment 3747553 View attachment 3747554

Denoising examples. It's probably the most important parameter in these.
"A * drooling dog, white in a forest" I think color correction was added though.
View attachment 3747623
Now this is what I'm talking about! Thank you for the tips and tricks, it's very helpful for establishing a scene.
 
Last edited:
I'm curious, what kind of prompts did you use? Especially for the hair to turn out so similiar
You can feed it photos. Give me something and I'll make one for you.

You iterate a lot of times and skim the good ones off the top.
You can also refine things you like.
Here's an attempt at making a kiwi girl
1666122543362.png
And here are her sisters when I ask for variations.
1666122584319.png1666122590395.png1666122597584.png

Well, someone did.
From 2017.
This video is 5 years old now
Two years ago
Two months ago
One month ago
NIGHTMARE NIGHTMARE NIGHTMARE
 

I love this series. The quality of shitposts has dramatically improved thanks to AI generated art in this very thread. Who would've made such an effort without it? I loled irl several times reading this thread. And people tell me this isn't art? It's the purest expression of human emotions. We enjoy looking at these creations. They make us happy. What's not to love?
 
My mistake, I was not aware that an analogy would be an obscure concept to you. Sorry if it puzzled you.
An analogy has to work logically for it to work at all. Your comparison implies you're too crippled to learn how to draw. The whole point of making one is that it's supposed to create a parallel.
Next time you want to put me in the deep fryer, I would appreciate it if you'd at least try to comprehend my actual meaning before you do it. That's all I have to say.
Next time you want to imply I'm too dumb to understand your meaning, maybe you should think up an analogy that works.
 
  • Autistic
Reactions: Horribadger
I think in the extreme people will still gravitate towards "artists" who curate and incorporate the generated art in a way they like. You still have to make a cohesive product people are willing to spend money on.

In the absolute extreme, everyone is isolated giving the AI their own prompts. Through the need for social interaction, they share what they've found to other people. It'd be like if we were all stuck in the Library of Babel but had internet access to share what we've found.
The key thing here is that the Library was useless when it was created and still useless now. What we need to build to make it useful, to find an actual (or at least plausible) cure to cancer and all the stuff that was hypothesised is we need a Librarian.

This is the #1 problem NovelAI has at the minute; it cannot accurately contextualise beyond one single scene. You can describe a basic setting, scenario, and a couple of characters, but it is hopelessly useless if you try and write a whole story in that manner, because there's no "constants" you can load into it to make sure it doesn't completely derail your story. Character A is supposed to be a young child, so don't write them into any mature scenes. Character B is a formless, eternal being, so don't describe it as having "arms" or any other mortal attributes. The AI will eventually fail, even if you give it the right info.

And that's why the Library of Babel is likewise useless outside of being a concept; you can search for a particular string (i.e. "The cure for cancer is...") and what you'll get is that string on one page, surrounded by 50 pages of complete junk, because that's just one permutation of uncountable millions. We need to make an AI that can actually store real information and produce documents with heretofore unknown information, based on actual physics/chemistry and synthesizing from real research papers made over the course of several decades. Then, when you ask for a book that has that aforementioned string in it, it's possible you may very well be presented with the real deal.
 
Back