/sdg/ - Stable Diffusion General

emma laugh at puny human edition

Old thread: >Starter guides
pastebin.com/raw/8SuS2LRy - Current main guide (Voldemort's)
github.com/sddebz/stable-diffusion-krita-plugin - plugin for Krita
rentry.org/GUItard - Former main guide (hlky's version)

github.com/hlky/stable-diffusion/wiki/Linux-Installation - Linux install script
rentry.org/ayymd-stable-diffustion-v1_4-guide - AyyMD Stable Diffuse v1.4 for Win10
rentry.org/sdamd - Updated AMD guide
Updating old installs

>Training
rentry.org/Stable-Diffusion-Training

>Compute Platforms with free GPUs
- Colab
- PaperSpace

>Colab with hlky's version
colab.research.google.com/drive/1AfAmwLMd_Vx33O9IwY2TmO9wKZ8ABRRa

>In-depth usage information:
wiki.installgentoo.com/wiki/Stable_Diffusion - Any Forums Wiki with tips
github.com/Maks-s/sd-akashic - Stable Diffusion Akashic Records
sgreens.notion.site/sgreens/4ca6f4e229e24da6845b6d49e6b08ae7 - gallery of artist styles
lexica.art/, krea.ai/ - prompt sharing sites

>Debugging
rom1504.github.io/clip-retrieval/
laion-aesthetic.datasette.io/laion-aesthetic-6pls/images
replicate.com/methexis-inc/img2prompt - SD Interrogator

>SaaS sites:
huggingface.co/spaces/stabilityai/stable-diffusion - Free txt2img
huggingface.co/spaces/huggingface/diffuse-the-rest - Free img2img
beta.dreamstudio.ai - Free txt2img, New accounts get 200 generations
pornpen.ai - Free txt2img, NSFW focused
dezgo.com - Free txt2img
huggingface.co/naclbit/trinart_stable_diffusion - txt2img specializing in animu

Related:
/h/ /sdhb/: /vg/ /aids/: /mlp/ /ppp/: Official discord: discord.gg/stablediffusion
NSFW discord: discord.gg/2ZhBzDGf8u
Tech support Matrix: matrix.to/#/#info:conlaboro.xyz
Stability.ai updates: twitter.com/EMostaque

Attached: 00019-3560892321-toned_emma_watson_muscle_princess_muscular_flexing,_brightly_colored_flowery_sexy_stripper_dress,_sexy_pear_hips,_full_body_port.png (512x576, 388.88K)

Other urls found in this thread:

image.pmgstatic.com/cache/resized/w936/files/images/film/photos/000/316/316902_e8e59c.jpg
1.bp.blogspot.com/-G7-Y471tySg/XmBlsVy6g1I/AAAAAAAADRI/NIj9KujLpvMO73HalfLjGecECa0NKRxWgCNcBGAsYHQ/s0/3.JPG
twitter.com/SFWRedditVideos

Attached: grid-0295.jpg (1536x2112, 478.66K)

old fag thread go!

Evidence of the furry art apocalypse

Attached: download (13).png (512x512, 532.64K)

furries run the world.

Attached: file.png (774x1286, 1.71M)

FIRST OFF FUCK EMAD AND THE SHIT HE CLAIMS

Attached: The Darkness.jpg (1920x1920, 2.32M)

exactly. That's a nigger elf, not a dark elf. Menzoberranzan isn't in africa.

Attached: download (12).png (512x512, 519.39K)

Attached: grid-0293.jpg (1536x2112, 450.62K)

Try jackals and dog breeds. This is Borzoi

Attached: download (16).png (512x512, 566.93K)

post link or it didn't happen

okay i've generated 150 emiliar clarke images at every reasonable age and it can't neoteny her at all. its like the model thinks she was always 25.

Is this the NAI finetune? If so, neat, but I'm not giving them my money. Non-local gen is yesterday's tech now.

Attached: werere.png (512x512, 393.74K)

First one is a hit.

>The Wicker Man 1973, attractive, sexy, dark, gloomy, desaturated, award-winning, by Greg Rutkowski

Attached: xy_grid-0001-3355929015.jpg (2122x4000, 1.52M)

Link to what, fag? This is standard-ass SD 1.4. No furry model needed, just promptsmithing. Furry art is gonna get flipped like a turd on a range

Attached: download (19).png (512x512, 540.45K)

I hecking LOVE THIS TECH!!!!!!!!!!!!!! I CAN'T GET ENOUGH OF IT ARHGGGGGGGGG

Attached: 1632666755880.png (1025x512, 808.7K)

Attached: grid-0297.jpg (1536x2112, 453.68K)

lol then nothing has changed. yiff in hell et al.

samurai furry?

Attached: Samurai.jpg (1521x2238, 1.09M)

Final furry.

Attached: download (18).png (512x512, 535.12K)

Man, that sexy keyword really kicked in later on those ancestral samplers, didn't it?

this got passed over when I posted it the other day so I'm gonna try posting it again. it's a very good prompt, high success rate:
>they dont make boobs like these anymore!!!

cherry-picked results from ~150 generations (of which ~80 were nsfw), 5 are unedited and 4 are lightly edited (regenerated same seed at different guidance to cover up an artefact or nipple): top center, mid right, bottom left, bottom middle are edited.

they're cherry-picked as I said before but the typical result was quite high quality compared to typical results for all the other prompts I've tried

Attached: they dont make boobs like these anymore!!!.jpg (1536x1536, 926.78K)

Some of those are pretty good. What prompt did you use? Try to get them with white hair and lips (impossible) if you can.

Oh so you're just cherrypicking like always. Which is great, but objectively this model sucks at furry overall. There are tiny little pockets that work well, but everything else is garbage.

Attached: erwereree.png (512x512, 344.03K)

how would I replicate a helmet like this in SD?
image.pmgstatic.com/cache/resized/w936/files/images/film/photos/000/316/316902_e8e59c.jpg
1.bp.blogspot.com/-G7-Y471tySg/XmBlsVy6g1I/AAAAAAAADRI/NIj9KujLpvMO73HalfLjGecECa0NKRxWgCNcBGAsYHQ/s0/3.JPG

Attached: file.png (640x640, 395.98K)

My guy, I do not need to cherrypick. I am a fucking promptsmith. I make my images one at a time with a deterministic seed. I can make whatever furry I want. The ultimate furry power is in the hands of a non-furry.

Attached: download (8).png (512x512, 506.48K)

The model is overfitted. you have to use [] in your prompts to tell the system to fuck off on the strength of the subject.

youthful child portrait of a young girl with blonde hair [[Emilia Clarke]]

Attached: 00013-1439144487-youthful_child_portrait_of_a_young_girl_with_blonde_hair_[[Emilia_Clarke]].jpg (512x640, 36.22K)

So I bought a used M40 that has 24GB, my main system is 3070TI with 8gb, how do i get torch or whatever the fuck its called to use 24gb instead of my main gpu. what are my max res size now?

>try to make image
>adjust prompt to more exact language
>adjust image weight
>change style of prompt
>adjust prompt to more ambiguous artsy language
>find something close and adjust seed
>adjust seed again
>and again and again and again
>give up
>come back a day later and try again
>adjust prompt
>get exactly what you want
Is there a better feeling?

Try hardsurface faceless helmet or something

Attached: anime, very detailed, long white hair, intricate dress, gorgeous face, beautiful eyes, pixiv, art by yoh yoshinari, kyoani, makoto shinkai.png (512x512, 420.96K)

I have a question about img2img
Whenever I manually drop a 512x512 image into it, I run out of vram.
But using the GUI to push the same image to img2img works just fine. What's up with that?

When you can use that seed and variations on that prompt to make 20 different things.

According to wotc it is.
So are orcs and recently they've added humanoid monkeys to the list.

Attached: anime, very detailed, long white hair, intricate dress, gorgeous face, beautiful eyes, pixiv, art by yoh yoshinari, kyoani, makoto shinkai (1).png (512x512, 450.43K)

Probably hardware acceleration.

What repo are you using? Hkpy one has a command line to choose the gpu to run

yea... rip d&d lmao

that shouldn't happen.. just tried it and i don't see a difference in memory usage between push and upload

are these images you generated earlier or are they from a different source?

Attached: anime, very detailed, long white hair, intricate dress, gorgeous face, beautiful eyes, pixiv, art by yoh yoshinari, kyoani, makoto shinkai (2).png (512x512, 423.11K)

can both gpus be used for the same task at once?

Try Babylonian somewhere in there.

I find that making variations is almost always a downgrade from the original.

I don't mean whatever some GUItard is calling variations, I mean being able to get images with consistent style and composition. Consider:

Attached: Cat.png (512x512, 468.45K)

Attached: anime, very detailed, long white hair, intricate dress, gorgeous face, beautiful eyes, pixiv, art by yoh yoshinari, kyoani, makoto shinkai (3).png (512x512, 452.26K)

And:

Attached: Dog.png (512x512, 482.3K)

I have a 3080 with 10gb memory and a 2080ti with 11gb, any difference other than the larger possible resolution at the cost of time spent per pixel? I'm using the 3080 and wondering if I should switch.

nope, not gonna make any difference, not worth the time

Attached: M0XAMYaVtMpbIQWJiFIDxshWMe+eR5CTCxEBEoAyw0BPgihd62IKQUFkoK0JCfmEEJERTIvncWznAWBSGDWMl7iMq+eyIpQ8DEFBYq1FpnKQOHY4Ppltwj+27zXC2TnAEQkwcVFiIpAkZOE4OgBC1aJyXzq9L+XxOAskXqwJw5AAAAAElFTkSuQmCC.png (512x512, 345.2K)

Anyone know if voldy's gui setup contain the neosecret optimizations by default?

Attached: arnold.png (512x512, 361.92K)

man the quality of the bate here sure got low.

Attached: Untitled-2.png (359x125, 14.02K)

Damn, white lips are hard.
"Close portrait of a white lipped black skinned female Elf with white hair, DnD, LotR, WH40K, lithograph by John William Waterhouse and Kyoto animation [[and Frank Frazetta]] and Greg Rutkowski"

Attached: grid-0298.jpg (1536x2112, 473.11K)

--medvram --opt-split-attention? it has that

It doesn't. You have to make your own edits to the py.