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

I will freak the fuck out if they make an AI that can make @AnOminous shitposts even better than I can.

Shit that's probably already possible.

Someone should actually do this. Then I can rest.
Could you explain to a moron who doesn't know shit about computers like me how you get these kinds of results? I tried a few on a whim and they looked... okay. Better than I could do certainly, but just kind of okay. How do you get from the first 'meh' iteration to something that looks good? Bonus points if it doesn't cost money.
People are actually good at this and probably put more effort into every image than the shit "artists" who are supposedly being replaced.
 
Could you explain to a moron who doesn't know shit about computers like me how you get these kinds of results? I tried a few on a whim and they looked... okay. Better than I could do certainly, but just kind of okay. How do you get from the first 'meh' iteration to something that looks good? Bonus points if it doesn't cost money.
I'm no expert and people probably have different and better processes than I, but here's what I do.

Without going into finer details and specific settings...assuming you're working on your own Stable Diffusion setup without paying for anyone else's, start by using a model that hasn't been trained on random garbage from DeviantArt. Depending on what was used in training, models vary in their strengths and outputs (as well was what prompts they recognize). If you want to mimic a specific medium (photography, watercolor, oil painting), include it as part of the prompt. If you already have an image with the particular pose or layout you desire, that can be used as a starting point with img2img. Otherwise, make a prompt, run it, and see if you're getting the desired things. If so, you're good, if not, adjust the weights of the prompts and the prompts themselves to get more outputs of what you want and less of what you don't want. Once you're getting the characteristics of what you want far more often than not, go all in and render a bunch of images instead of just tiny batches for testing purposes.

Go through that collection and you'll probably find a few that are really close to what you want. If it's the desired pose/layout, go to img2img and generate more images that will be not unlike your input image. If it's smaller things (like wrong number of fingers/undesired object/clothing) and you like the majority of the image, go to inpaint and select what you want to redo, then it's time for another big batch. Pick the desired or best output, then inpaint again. And again. Being comfortable with photoshop helps, as you can fix smaller flaws in that instead of inpainting.
I want an image of royalty in armor with long billowing hair. After playing around with the prompts, I get this.
00147.png
It's missing a few things I want, but it's a good starting point. On to image2image mode.
00063.png
That's better but also worse; I want her to be wearing fantasy armor that's not purely eyecandy. Let's fix that. Time to inpaint.
00088.png
Much better, but still need to fix that cloud in the background where her tits used to be. Inpaint again.
00116.png
Every time I mess around and experiment is still a learning experience.
The only real limit is your imagination and the time you're willing to invest. For specific online services, I have no advice other than don't use DeviantArt's service because it's fucking terrible.
 
The biggest danger I see in technologies like these, and I posted it before, is to get into a situation where the entire power of it lies in the hands of a select few. I wish these retards could understand that they might be playing directly into the hands of these select few, which will be a lot worse than losing some money from some footfaggot because the AI generates cheaper feet.
It never ceases to amaze me that people never realize how bad this could get when it's left . This Pandora's Box opened a quite a number of years back, and YOU CANNOT STOP IT. There's way too much stuff out there now.

It's much better that everyone knows about this and knows about how much shit can be potentially faked so the power of this technology gets hampered by an informed populace and cultural safeguards that can follow.

:optimistic: I know. But treating this as a part of your arsenal will probably work much better than trying to limit access to it.
 
This is about a week late, but that whole thing equating AI with plagiarism cause they can cause it to crash by flooding ArtStation with No Smoking AI art signs really misses the target by a mile.

Unless they can find direct lifts/copypastas, and I'm yet to see one, this is no different from looking at an artist you admire and then emulating his/her technique/style. Congrats, you got a stupid software to crap out by fucking with its data set, but that doesn't prove the AI copypasted your work.

And even then, since none of these are straight up rips, an argument can still be made about the product being transformative. A number of artists I know IRL are also music guys and they have LOTS of tracks made with samples (good sounding ones at that). Oh, and let's not forget about the amount of people who photobash the hell out of their work, and the biggest and loudest advocates are industry big times that work at major VFX studios and AAA game studios.
pretty much all art and music is at a point now where its gonna look or sound similar to someone else. songs will sound the same because they use similar notes. some art looks the same because they use the same style. look at the rise of the cal arts style in cartoons for proof of that. does that mean its stealing if 2 cartoons have the same animation style? or did both animators simply go to the same school and learn to draw the same way? even the images i have seen that are copies of other artists art have their differences.

you can argue that AI art is bad because its steals business from real artists all you want but you gotta keep in mind that the odds of whoever is buying the AI art instead of art from an actual person actually caring enough about art to bother buying from a real artist are pretty low. like there are gonna be people who buy computer generated art because they like it more than human made art. thats just a preference thing and nothing will change that.

any real artist knows that art is subjective your style might not be everyones cup of tea. but someone else can draw that same thing and people love it. thats just how art is. its always been that way and it will never change. some people are gonna like AI art more than human art and some will like the opposite. but getting upset because a robot or whatever else is more popular than you just comes off really petty and shows you care more about your art being popular rather than actually making said art.
 
This is definitely..... original.

FB_IMG_1671695072154.jpgFB_IMG_1671695062689.jpgFB_IMG_1671695079838.jpgFB_IMG_1671695083833.jpgFB_IMG_1671695716428.jpgFB_IMG_1671695052073.jpgFB_IMG_1671695056023.jpgFB_IMG_1671695068206.jpg
Screenshot_20221222-014516.png

Somewhat cursed they may be, but at least it blows chunks out of the argument that AI artists lack the ability to create meaningful juxtapositions which may pose deep questions, to design entirely unique compositions, and transcend design conventions for almost any subject.

Drawslaves probably have wet dreams about coming up with something that original & unique; but unlike their nemeses, they can only crank out endless recognizable variations of the same current-thing cartoon characters, movies, or cat-girl/fox-trap hybrids as donut steel OCs, while using the same half-dozen or so facial expressions & poses.

And now, for something completely different....
 
Last edited:
but getting upset because a robot or whatever else is more popular than you just comes off really petty and shows you care more about your art being popular rather than actually making said art.
What they're mad about is it also shows they suck at art, literally a mindless robot can make better art than most of these fartistes.
 
It's usually the trad/digi arts types that are MATI about AI art. These porcelain guns are something that you could probably see at a modern art exhibition, and is something I'd actually like to see instead of just another labelled boulder or picture frame waterfalls.
Porcelain guns are the kind of art that modern art as a concept was meant for, boulders and blank frames are what it's used for.
 
Could you explain to a moron who doesn't know shit about computers like me how you get these kinds of results? I tried a few on a whim and they looked... okay. Better than I could do certainly, but just kind of okay. How do you get from the first 'meh' iteration to something that looks good?
I think others explained the process pretty well, but @AnOminous is right that you have to put in a lot of work trying to make it look good. I don't think the technology is quite at the level where you can just use most of the images as-is. For example, let's look at that first picture from my earlier post, but before I did anything to it.
jonas1.png
We have two major problems: Firstly, his hands are atrocious. Gotta fix that before we fix anything else. So we generate some more pictures using the exact same prompt until we get some nice looking hands.
fixed hands.png
Those look nice, so we get rid of the old bundle of sausages and put these on.
Now, the next main problem: the goat. Right now it looks more like a miniature horse, rather than a something deserving of the name. So we do the same thing again...
goat fixed.png
Eh, it's okay but one of the horns is cut off by the top frame. Still, a hell of a lot goatier than it was. Stick it on there.
jonas.png
In the end, you get something that looks a hell of a lot better. Nice picture of a medieval peasant posing near the river.
But hey, let's expand things. Using another program we can do inpainting and outpainting, so we can expand the frame until we get something like this:
jonas full frame.png
There we go. Has a nice mix of color, opens things up a bit. It's not perfect, but it's good enough for me.
Keep in mind, all of the editing was done with Photopea because I'm too lazy to pirate Photoshop or even download GIMP. Anyone can do this given enough time, and someone skilled could do it much better.
elf.png
Anyway, good luck on making this kind of art in the future. Keep at it and you can really make some great looking stuff.
 
Last edited:
Decided to do a whole lotta stuff, basically made some mock up hastily generated fake Sci-fi anime mecha series (also some bonus stuff) :diddler:
 

Attachments

  • 00372-2780070579-1girl, short spiky orange hair, brown eyes, wearing white leotard open down t...png
    00372-2780070579-1girl, short spiky orange hair, brown eyes, wearing white leotard open down t...png
    591.4 KB · Views: 57
  • 00401-2020432246-1man, blonde hair, wearing cyberpunk facemask, long black trench-coat, with b...png
    00401-2020432246-1man, blonde hair, wearing cyberpunk facemask, long black trench-coat, with b...png
    677.6 KB · Views: 48
  • 00413-1991983976-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, bla...png
    00413-1991983976-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, bla...png
    799.5 KB · Views: 44
  • 00414-724492401-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, blac...png
    00414-724492401-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, blac...png
    790.6 KB · Views: 45
  • 00420-1013886763-1man, wearing red 1920's sci-fi body suit, short black hair, in post apocalyp...png
    00420-1013886763-1man, wearing red 1920's sci-fi body suit, short black hair, in post apocalyp...png
    643.5 KB · Views: 44
  • 00386-3798384417-middle aged woman, wearing cybernetic costume, black hair, sinister facial fe...png
    00386-3798384417-middle aged woman, wearing cybernetic costume, black hair, sinister facial fe...png
    512.5 KB · Views: 48
  • 00389-3531639638-middle aged woman, wearing cybernetic costume, black hair, sinister facial fe...png
    00389-3531639638-middle aged woman, wearing cybernetic costume, black hair, sinister facial fe...png
    614.6 KB · Views: 47
  • 00409-2354815745-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, bla...png
    00409-2354815745-1man, bald, cyborg body and face highly detailed, red eyes, robotic arms, bla...png
    727.3 KB · Views: 51
  • 00034-2506841969-90's anime art style, official art, black hair, wearing school uniform.png
    00034-2506841969-90's anime art style, official art, black hair, wearing school uniform.png
    601 KB · Views: 66
  • 00038-700653067-1girl, 80's anime art style, official art, black hair, wearing school uniform.png
    00038-700653067-1girl, 80's anime art style, official art, black hair, wearing school uniform.png
    544.8 KB · Views: 57
  • 00063-1879910456-male, bishonen, long brown hair tied into a ponytail, wearing a seifuku.png
    00063-1879910456-male, bishonen, long brown hair tied into a ponytail, wearing a seifuku.png
    419.1 KB · Views: 47
  • 00087-2559985567-1girl, short blonde hair, blue eyes, wearing ugly Christmas sweater with blue...png
    00087-2559985567-1girl, short blonde hair, blue eyes, wearing ugly Christmas sweater with blue...png
    552.9 KB · Views: 55
  • 00092-2896117917-1girl, short pink hair, blue eyes, wearing ugly Christmas sweater with blue j...png
    00092-2896117917-1girl, short pink hair, blue eyes, wearing ugly Christmas sweater with blue j...png
    642 KB · Views: 56
  • 00147-3273994326-_a comic potrait of a female necromamcer with big and cute eyes, fine - face,...png
    00147-3273994326-_a comic potrait of a female necromamcer with big and cute eyes, fine - face,...png
    514.8 KB · Views: 48
  • 00207-3310148430-1girl, short black hair, wearing red Letterman jacket with patches with blue ...png
    00207-3310148430-1girl, short black hair, wearing red Letterman jacket with patches with blue ...png
    579.1 KB · Views: 47
  • 00290-1356311841-1girl, wavy black hair, blue eyes, wearing plaid dress in a cafe, masterpiece...png
    00290-1356311841-1girl, wavy black hair, blue eyes, wearing plaid dress in a cafe, masterpiece...png
    508.9 KB · Views: 45
  • 00425-78973296-1man, wearing viking armor, wearing a fur cape, standing in the snow, blonde ha...png
    00425-78973296-1man, wearing viking armor, wearing a fur cape, standing in the snow, blonde ha...png
    606.6 KB · Views: 40
  • 00434-1019577657-1man, wearing samurai armor, with samurai masked helmet,  highly detailed, re...png
    00434-1019577657-1man, wearing samurai armor, with samurai masked helmet, highly detailed, re...png
    704.6 KB · Views: 43
  • 00438-607955295-1man, WW2 combat gear.png
    00438-607955295-1man, WW2 combat gear.png
    563.3 KB · Views: 47
  • 00451-628671182-1girl, wearing WW2 combat gear, highly detailed, hands on hips, short brown ha...png
    00451-628671182-1girl, wearing WW2 combat gear, highly detailed, hands on hips, short brown ha...png
    559.1 KB · Views: 59
I will never understand how anyone can justify this whole consent thing with regards to training data. It is not copying their work, it is taking inspiration from their work. Stanley Kubrick doesn't have a patent on people staring at the camera, nor can he stop another director/of photography from using that.
Why anon enjoys artist tears:
1671857556783672.png
I'm hopeful this guy's story book really did happen and is not just some hypothetical being spun to cause some MATI. That's an actual use case for shit like this. It's power to the people. Economic barriers are reduced and better things are created.
 
I like that AI chads are calling virgin artists "paintpigs."
Really? It's sort of a forced meme from a fake news article a /pol/tard posted on IQFY. It's a little funny but meh, IMO.

Also I never thought about this issue directly and I hate the use of the word "folx", but this is a valid point:
1671717115563941.png

Reddit chimes in with similar stories:
1671770697496489.png
Another one:
1671770621595232.png

More discussion here:
 
koboldai was cool until every AI dev cancelled their colab subscription and started using up all the free TPUs to generate horrific "art"
my gpu cant run koboldai and my cpu shits itself whenever i try so it sucks
 
  • Feels
Reactions: Mr.Miyagi
koboldai was cool until every AI dev cancelled their colab subscription and started using up all the free TPUs to generate horrific "art"
my gpu cant run koboldai and my cpu shits itself whenever i try so it sucks
Have you tried NovelAI as an alternative? I enjoy it.

Also, found this on /pol/. Karla Ortiz (the GoFundMe lady) and her followers on Twitter are insane:
everyone_please_repo.png
 
Back