/sdg/ - Stable Diffusion General

Old Bread: >Local install
GPU: rentry.org/voldy
CPU: rentry.org/cputard

>AMD GPU Guides
Native: rentry.org/sd-nativeisekai
Docker: rentry.org/sdamd
Onnx: rentry.org/ayymd-stable-diffustion-v1_4-guide

>M1 Supported Fork

>Cloud Hosted Install
Colab: colab.research.google.com/drive/1Iy-xW9t1-OQWhb0hNxueGij8phCyluOh
Paperspace: rentry.org/865dy

>SaaS Sites
txt2img: huggingface.co/spaces/stabilityai/stable-diffusion | beta.dreamstudio.ai | dezgo.com | (NSFW) pornpen.ai
img2img: huggingface.co/spaces/huggingface/diffuse-the-rest
inpainting: huggingface.co/spaces/fffiloni/stable-diffusion-inpainting | inpainter.vercel.app/paint

>Krita Plug-ins

>Other Guides
Inpainting/Outpainting: rentry.org/drfar
Upscaling images: rentry.org/sdupscale
Textual inversion: rentry.org/textard
Training: rentry.org/informal-training-guide

>In-Depth Usage Info
Any Forums wiki: wiki.installgentoo.com/wiki/Stable_Diffusion
SD Akashic Records: github.com/Maks-s/sd-akashic

>Tools & Resources
Big index: pharmapsychotic.com/tools.html
Artist styles: rentry.org/artists_sd-v1-4 | urania.ai/top-sd-artists | sgreens.notion.site/sgreens/4ca6f4e229e24da6845b6d49e6b08ae7
Prompt sharing: lexica.art/ | krea.ai/
Dataset: rom1504.github.io/clip-retrieval/ | laion-aesthetic.datasette.io/laion-aesthetic-6pls/images

>Finetuned Models

/h/ /sdhb/: /vg/ /aids/: /mlp/ /ppp/: Official: discord.gg/stablediffusion
NSFW: discord.gg/2ZhBzDGf8u
Training: discord.gg/xVsyrmhQWS

Stability.ai updates: twitter.com/EMostaque

Attached: fsekj.png (512x512, 715.63K)

Other urls found in this thread:


Is there a trick to outpainting with Voldy's "poor man's outpainting"? I seem to only get a huge black border around the original image.

We need AI to train on more artists like Jeremy Geddes, he can do a lot of dynamic poses, AI really responds to that too. More of that please, whoever can train it

Pretty good guide

>Native: rentry.org/sd-nativeisekai
this should be under local install CPU
change in next op

Sounds like a bad denoising value but outpainting in general needs work

Fucking nice
I just added red hair ahead of that and it already looks good

Attached: 00202.png (1024x1152, 1.12M)

Is anyone else using the stable-diffusion-intel version from the AUR? Any guesses how to get the webUI?

how do I turn this frogs into a bunch of other stuff like fruitsor cables and shit

I have SD and all the extras, with a few Textual Inversion models . But I dont know the settings to turn images into other images that look like it

Attached: 1654510028217.png (855x684, 79.48K)

Attached: 17937-2752246837-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (512x512, 411.73K)

Consider the following.

Save your fucking prompt and seed if you in have any hope at all of upscaling your photo with SD.

Attached: kaiba.png (768x448, 448.66K)

Attached: 00907-552507460-teenage princess, (cute nose), bright playful eyes, (detailed) pupils, (sexy pout), dynamic posing, ((perfect face)), Albert Lyn.png (512x768, 620.22K)

Based pooh-bear remixer posted a guide in the last thread.

auto-upscale on voldy when ?

Attached: 17955-1676150792-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (512x512, 391.47K)

how do you guys make Elsas look reasonable?
is it just a matter of keeping the image in the oven for longer?

Attached: file2.png (512x576, 477.07K)

> "It's 0400 ChronoSpex, user, time for your geneseed extraction"

Attached: 04227-6004-AdMech.png (448x576, 386.12K)

>I waited 20 minutes for this
I'm sorry, stable diffusion is horse shit

Attached: a_plate_of_green_spaghetti_with_sauce__photograph__aa96e4d3.png (512x512, 297.41K)

looks like Waifu Diffusion does not know (Cardcaptor) Sakura and just makes some kind of palette swapped (mostly red) Cirno instead

Attached: grid-0037.jpg (3072x3072, 1.2M)

Attached: 17958-2364791949-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (512x512, 355.62K)

What's the best resolution I can expect with a 3060 ti?

Attached: 1488187721889.gif (413x233, 2.62M)

It's all about artist tags and attributes. The actual content of the photo matters little in comparison.
I dare OP to use the same attributes and style and just change Elsa to an attack helicopter. It would look roughly the same, just an attack helicopter instead of Elsa.

teenage princess man, this is a fucking great prmpt
Looks awesome user, 10/10 would battle against

Attached: 00205.png (1024x1152, 994.39K)

This is FUCKING AMAZING. I haven't been this exited about using something in years.

Haven't updated voldy in a while.
Now it has launch.py? Webui.bat still works for launching, but launch.py starts to install torch. Does it now install shit on the system and not in venv folder?


Attached: 17960-2181138963-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (512x512, 404.71K)

Attached: download (15).png (512x512, 395.25K)

For the user that requested that I add Spider-Man.
Sadly, this is the best img2img could do with a prompt.
I'm don't really know how to use img2img. :(

Attached: 00058-3404936075-((Frank Cho)). Dry acrylic painting._((Spider-man)) holding a camera. High detail. Intricate..png (640x704, 815.64K)

Where is
github.com/cmdr2/stable-diffusion-ui could get to 1024
My 2080ti does higher its just not reliable probably 1280x1024 absolutely max I wunna get a 4090ti for 2048x2048

based sakura enjoyer
you should try kinomoto_sakura since that's the tag on danbooru iirc

Attached: 00612-3275629728-(Alphonse Mucha), ((Frank Cho)), (Donato Giancola), [Thomas Kinkade]. (Platinum Blonde)), Thin Black choker. (((Firm big breast.png (448x640, 421.02K)

Attached: mobo.png (512x512, 297.53K)


Attached: kitty.png (512x512, 530.27K)

yeah it really look nice

Attached: 00905-552507458-teenage princess, (cute nose), bright playful eyes, (detailed) pupils, (sexy pout), dynamic posing, ((perfect face)), Albert Lyn.png (512x768, 608.56K)

Attached: 00178.png (1024x1024, 865.79K)

I did try both kinomoto_sakura and cardcaptor_sakura (and both together) with no luck. It's probably not in the training data at all

> "Just try it I'm behind 7 layers of Terminator Armour"

Attached: 02295-1075-AdMech-Terminator.png (512x512, 437.07K)

what was the reason of this software? was there any? or just some dude wanted to create this from the get go?

> what was the reason of this software?

Attached: Anime_Princess_By_Noriyoshi_Ohrai.png (512x658, 593.13K)

How do you get these great results. Mine are all bad

Attached: 1662578056596526.jpg (976x850, 57.91K)

Can the based nudeshopper from last thread explain his settings? (ie which diffuser, what denoising range to choose etc.)

Attached: 17974-1220838479-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (1088x1088, 1.81M)

daily armpit

Attached: 1663330260439-5.png (512x1024, 770.88K)

Is there a better colab for voldy than the one from OP? Im so sick and tired of only being able to create one picture at a time. Its been like that for the past 6 days...

if everything works out for him, he will get some billion dollar deals


Attached: 04106-3642975628-2girl, kiss, Makoto Shinkai, Kyoto Animation, Pascale Campion, (Franz Xaver Winterhalter), Kentaro Miura, Lilia Alvarado, (Ayami.png (512x512, 380.07K)

puting a lot of work to zero in on desired parameters, good usage of prompts, trail and error

a lot of them are actually bad

Attached: Sans titre.jpg (3437x1786, 2.02M)

>no abs

We make 100 at a time, pick the best

Attached: download (93).png (512x512, 266.75K)

Port voldemort's UI to colab then. All it needs is an input form bound to some vars.

and iterative editing in img2img

Attached: undress.png (2560x581, 1.89M)

proooompt swap?

bust shot, evil monster ((bloody sinew spinal hair)), wet, ((torn flesh clothes)), (wide_eyes) (eyeliner), underwater , ((athletic, muscled)), geometric runes, intricate, elegant, glowing lights, dramatic lighting, blood teeth gore organs flesh guts offal bones, by greg rutkowski, Franz Xaver Winterhalter, octane render, ((bathtub background)),kneeling in water

Negative prompt: ugly, fat, obese, chubby, [blurry], bad anatomy, disfigured, poorly drawn face, , (ugly), (poorly drawn hands), messy drawing, penis, nose, eyes, lips, eyelashes, text,red_eyes, large_breasts

Steps: 28, Sampler: Euler a, CFG scale: 8.5, Seed: 3458194830, Size: 512x512

Attached: 17975-3458194830-bust_shot,_evil_monster_((bloody_sinew_spinal_hair)),_wet,_((torn_flesh_clothes)),_(wide_eyes)_(eyeliner),_underwater_,_((athlet.png (512x512, 420.08K)

I've already said that CF is good at fixing eyes but generally makes them darker, so if you want blue eyes, set the prompt for green, which CF will then turn blue
Apparently it understood green jacket instead of blue eyes here though
Copy the prompts that people that get good results are using

Attached: 00227.png (1024x1152, 1.05M)

Attached: 1660160687215.jpg (2048x3072, 375.1K)

good taste

Ok, WD decided that green is for backgorund if it's in the prompt

Attached: 00228.png (1024x1152, 1.08M)

In the past 2 days I have generated around 1000 sexy Elsas. Not all of them are nice looking, but with each iteration the prompt gets better

Attached: Screenshot at 2022-09-16 16-44-41.png (1605x844, 768.78K)

Use fill for masked content and high number of steps, expand in one direction at a time. Outpointing could be improved greatly with good noise initialization. I have it on my todo list.

Oh ok thanks makes sense.
Thanks, i will definitely try some

This defintely reminds me of some actress
Still using green for stuff that isn't eyes despite
>bright green playful eyes
Which it weird because
>bright blue playful eyes
works as intended
Fuck it's really like trying to talk with a child

Attached: 00229.png (1024x1152, 1.12M)

There's nothing I'd do different. It clones the repo, installs requirements and loads model from drive (I changed that). Theres another voldy colab (colab.research.google.com/drive/1zN99ZouzlYObQaPfzwbgwJr6ZqcpYK5) that has the exact same bug

I was so excited about this tech and I really got onboard, generating landscapes and fantasy cities and comfy wallpapers, but within a week these generals just totally devolved into manchildren constantly just constantly generating big tit anime women on their GPUs and thats just extremely annoying, sad, and depressing and I am no longer interested because there is no longer anything interesting or unique in these threads. Its just bigger tits every day and more prompts on how to get them.


Attached: 00000-~1.png (512x768, 826.04K)

I don't think you should be using ugly and disfigured negs bro


seething arttroon

what if you described her more? Like add pig tails or short twintails. At least you might come out with a knock off sakura

Ugggh bigger

>I was so excited about this tech and I really got onboard, generating landscapes and fantasy cities and comfy wallpapers, but within a week these generals just totally devolved into manchildren constantly just constantly generating big tit anime women on their GPUs and thats just extremely annoying, sad, and depressing and I am no longer interested because there is no longer anything interesting or unique in these threads. Its just bigger tits every day and more prompts on how to get them.

Attached: 1663172322494694.png (512x512, 272.27K)

Classic S𝈨𐌴 moment

>within a week
stop complaining and post your art then, newfag