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

  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
Sinner's Sandwich said:
I really don't think so. Quite the opposite. People would think you are one of these weird history nerds learning stuff people in ancient times did.
Click to expand...
In other words, a literal ableist Nazi murdering people.
Allanon said:
Stable Diffusion is for photorealism and paintings and it is very bad at drawn mediums like anime unless you put in specific modifiers like artist names and styles. It is currently uncensored and free.
Click to expand...
So far I've had the best luck with SD despite really having no clue what I'm doing. 1660 6 GB.
 
AgendaPoster said:
I don't like where this is going.
Most of you nerds are using this to generate females and obvious masturbatory material.
this shit should be haram and you should be pushed by society and everyone around you to chase real females, in real life, in clubs, on the streets, at the job and everywhere else, instead of using your GPUs for fucking anime waifus.
Stop being such fucking weebs and make an effort to wake up
Click to expand...
But I'm having so much fun.
311430044_659286345589445_5590194857702246498_n.png 312350959_1088531205160025_5321507601425586362_n.png
 
So what have you guys made during the downtime? Also anyone experimented with merging models? Right now I'm using 60% waifu diffusion with 40% SD 1.5 and the mse autoencoder
plebbit auto.png
Stolen from plebbit
The WD model is trained on a much wider variety of artists, I feel that it makes up for some of what regular SD is lacking.
(All of the below come from img2img, I find it the most fun to play around with)
02209-93447486-A single very fat man with round glasses, screaming. ((he is sweaty and has bad...png 02451-2921591978-pudgy man pretending to be a woman with long hair, poor hairline and dead sta...png 02712-3362476850-a single dog, detailed red eyes, slobber running down jaw, digital art, (shar...png 02689-1425616851-a red octopus, highly detailed, digital art, (sharp focus_1.2), intricate, h...png 02589-1521656602-A red octopus, breast, highly detailed, digital art, (sharp focus_1.2), intri...png 02430-1288809329-Soyboy with a headset making an _O face, MOUTH WIDE OPEN, thin mustache, brow...png
 
tmp_nds3puc.png

Mah mod is fuckin' my wife, I CAN'T DO IT I CAN'T TAKE THIS SHIT NO MORE *dog whimpers*
 
I was bored and built a hypernetwork while I couldn't shit post here.
Holy fucking cow-rimming furtranny. This stuff is creepy. You take around twenty pictures of someone's face and you're able to plaster it to anything after enough training time. Deepfake? Nah, that is so 2020 - Just use a day to fine-tune the AI and you can insert yourself discussing the final solution with Hitler at the Wannsee. Soon you can't trust anything anymore if you can't see it with your own eyes in reality and you can touch it with your hands.
 
Is there a checkpoint that isn't anime and doesn't murderrape faces? Trying to trick a merged realistic and anime checkpoint seems to be the best I can do right now. Realism might be asking too much at this point.
 
Most of what I did during downtime sucks, but here's some not shit stuff.

Aoi WIP 1.pngRose WIP 2.png

This one has some big problems, but eh, I like what it did with the clothing here.

Frost WIP 1.png
 
Last edited by a moderator:
Gaunt King Ithan Rilph said:
So what have you guys made during the downtime?
Click to expand...
Fucking nothing because my laptop melted. Not directly due to S-D (it was a recurring power issue) but it probably didn't help.
But holy shit I don't hate being forced to upgrade. I'm looking forward to applying skills I had to develop on prompts that took 50 seconds to gen that now take 2. Or big DPM Adaptive gens that took 10 minutes being done before I'm finished tweaking the next prompt.

I was already addicted to X/Y plots for sampler/parameter hunting that took hours, being able to iterate models/hypernets in the time it takes to go for a shit is mindblowing.
 
Puff said:
Is there a checkpoint that isn't anime and doesn't murderrape faces? Trying to trick a merged realistic and anime checkpoint seems to be the best I can do right now. Realism might be asking too much at this point.
Click to expand...
A lot of it is finding the right art form for the right thing. I use tags for 3D rendering, then sometimes for sketches, and other times for digital paintings. Also getting the right artist tags is important.
Use lexica to help you find artstyle prompts. I recommend merging SD with WD in the 60/40 range in favor of either one.
 
Involuntary Celebrity said:
here u go bro, threw in a kasumi too
View attachment 3767872
Click to expand...
Misty looks fine but pikachu looks like he's beginning to melt. When I first tried to do pikachu I checked /vp/ to see if they had anything on AI generated pokemon, and to my surprise there was... for the girls. I think there was a buff ash mixed in but that's not helping. All I wanted was some cute images of pikachu, and all I got was schizophrenic scribbles and coom.

On a normal note and relevant to my interests, someone trained what I presume is a hypernetwork, on Team Fortress 2 environment concept art. Twitter thread (a).
Some of the results
1667276904124.png 1667277044079.png 1667277217364.png
And similar official art for reference
1667277129587.png 1667277098119.png 1667277182938.png
 
You must log in or register to reply here.
Back
Top Bottom