No.63382
>general
go away
No.63430
>>63372This is a really nice, comprehensive, organized collection. Thanks OP. What have you been using diffusion for?
No.63436
there's always something uncanny about the results i've seen in 4chan threads. like they're not bad, but they need further tweaking by a human artist
No.63453
>>63436some of them look amazing though. I think the ones that look really good probably have guys using tools to correct shit not just taking the raw output. You can still use the AI to do it instead of shooping it by hand.
No.63454
>>63436the actual porn is pretty generic in pose from what I've seen. Only the body shape seems to be any good
No.63455
>>63454you can just give sd whatever pose you want with img input
No.63465
>>63464are you the wiz that made the world building thing where it was normal for succubi to have huge breasts or something like that, and they had mixed asian/russian genetics or something? if not, you might have an equal lurking somewhere
No.63466
>>63465I'd be pretty astounded if he was a different person.
No.63467
>>63465No I am not that wiz, but I remember him. I get second hand embarrassment thinking about his weird fetish world. I think he was in the fap thread or the draw thread recently talking about paying people to draw big boobs on to old Hanna Barbera cartoons or something.
I simply enjoy loli and I enjoy big boobs, and sometimes I feel it is nice to combine the two. If robots can do that, then maybe robots aren't so bad after all.
No.63468
>>63467fascinating. thanks for replying. im also becoming very fond of stablediffusion and ai generated stuff. until they corrupt it with ads, weird biases, and other ways to manipulate the user, the interactions with these machines feel pure and good somehow
No.63648
>>63468They already censored it due to people creating child porn
No.63652
>>63648what do you mean? the program itself doesnt censor anything, you can put in whatever you want. the default model might not have been trained on what you input, for obvious reasons they. but theres other models where ppl train them off porn and hentai and mix them up. if you mean services like midjourney or whatever, yeah that's their choice, it is their hardware running the program they can decide what content they want being created. you are free to just run the programs yourself tho
No.64710
I will give this a shot later, usually I'm too retarded to get stuff like this running.
No.64711
>>64710https://yewtu.be/watch?v=onmqbI5XPH8Follow this tutorial closely and it should all be ok. I had problems a couple of things, but I think that's because I dick about with my computer settings too much. If you follow this one, make sure to download 3.10.6 python, because it doesn't work on newer or older versions, surprisingly. And when installing python, make sure to check the box for adding a path variable. If you get stuck on a step make a post here and I or someone else might be able to help get it started. It isn't brilliant with human figures and faces, but for actual artwork I can see why all the web artists are frightened.
I can't stop making loosely inspired Gerhard Richter seascapes at the moment.It even gets the canvas effect right.
No.64716
>>64711do not ever use software that
-is not a self-contained binary
-does not have a simple one click installer
-requires techbro youtuber faggots to explain how to install and use the thing
get a gui for stable diffusion. i personally use 'NMKD' because it takes care of installing stable diffusion and getting all the bullshit it needs, it does it all for you. but any gui or frontend for sd is better than nothing
when you eventually learn enough about how to run and generate images, next you will want to experiment with different models. go to
https://civitai.com/ to find stuff
unrelated but here is an inpainting demonstration using nmkd. you select the area you want to replace, see what generates. when it doesn't look right you crank down the image influence and let it go wild and basically RNG something aesthetic. when you get something good, you can raise the influence just enough so the image remains visually constant but there is enough variation to allow you to see any possible improvements. when you find something good, you use that as your new starting image and adjust the mask to fix any areas that aren't good. you do this repeatedly until the whole is something you are ok with. this is my process at least
No.64719
>>64716Never use software that
-isn't FLOSS
-you don[t hage to compile for yourself
-requiresmyou to type ~/configure make make install
-has less than 15 dependencies
-doesn't require specific versions of some dependencies
-you can install without having to modify the makefile
-takes less than a day to compile with 4 processors
-doesn't throw warnings throughout compilation
-doesn't require rust
-doesn't require dbus
-doesn't require systemd
-comes with proper documentation
No.64723
>>64719FLOSS is cool but i like the unlicense better, all the rest of that is actual aids hell
No.69857
>>63372unpopolar opinion
NovelAI 4.5 > alle
No.69871
>>69868YUP, although
>NovelAIto avoid multiplying threads
No.69885
>>69871I use Illustrious not NovelAI.
No.69887
Stuff like Fubo_XL, Anime_XL3, Kioshi_3 allows me to use and mix various styles of various "artists"
however, you can use "anime screencap" tag to recreate some character's initial anime look.
No.69923
For anyone wondering: you can work with novelAI in two ways:
1. "An image of …, wearing…, with …, "such" pose, masterpiece, "such author" style"
2. Throw some *booru tags: "blonde hair, lond hair, very long hair, smug, grin, looking at viewer, …"
No.69938
>>69935thanks!
it uses "yukkuri shitteite ne" tag and the name's Inubashiri_momiji