/sdg/ - Stable Diffusion General

Attached: 120.png (1024x1024, 1.06M)

Attached: 125.png (1024x1024, 998.6K)

For me, it's the simple things.

Attached: 1663294609699.png (512x640, 518.41K)

pls give info on prompt keywords for poses

Attached: h4545h.png (448x640, 361.74K)

AMDbros please why wont pytorch build

Attached: file.png (900x1493, 381.56K)

Why is /sdg/ obsessed with Nikocado's ass?

none of these sexy, busty, hijabi women exist, crazy

In a previous thread someone was interested in movements and actions from the character, I tried different verbs with a simple prompt here and it seems to work well, some seeds showed mostly the same though

Attached: xy_grid-0019-2411442676-picture of a fit woman, wearing jeans and tank top, facing the camera, full body, exquisite hands, good details, long hair, zoom.jpg (4000x577, 326.49K)

Fix aspect ratio noob

Attached: 00512-400472906-Alphonse Mucha, ((Arthur Streeton)) and so-bin. Princess Zelda from breath of the wild. Firm breasts. Midriff. ((Abs)). Wide hip.png (512x512, 352.69K)

Attached: beautiful_japanese_chawan_beautifully_artistically_on_wooden_table_in_a_home,_close-up___shot._art_by_Hiroshi_Yoshida,_jean-leon_gerome,_Jakub_Różalski,_Hubert_Robert,_Albert___Goodwin,_volumetric_light,_octane_render.jpg (788x926, 135.03K)

that's just a neat little picture, wtf is wrong with you???

Oh man, this is going places.

Attached: gnome7.png (512x512, 374.76K)

Someone trains the AI at pepes please

Attached: Kaktuslutscher.jpg (750x697, 366.19K)

That’s a tree diffracted through a water droplet, you homo.

do it yourself

Also, a tree's already been done, disqualified for unoriginality

prompt was "haunted house on a hill"

I keep getting black squares instaed of pictures

no, I am too retarded

thanks user

try to get a specific scene painted on it using inpainting.

Because they’re all cocksucking subhuman faggots

Is there any benefit to training versus Textual Inversion? And have any anons tried doing Textual Inversion with more than 5 pics? I feel like 3-5 may be too little for SD to understand stuff.

Why are all of my results so fucking shit. What am I doing wrong?

Emad is a Giga Chad

Attached: 00643-3603481069-(soyface), ((wojak)), ((indian)), (muscular), (manly), jew.png (512x512, 280.14K)

It’s insane. I guess this is the most ethical way to fap to a hijabi gal without defiling her purity bestowed unto her by Allah.

check the guide

try messing with

Attached: 1656120895045.png (377x81, 6.92K)

skill issue

Attached: 07744-3860293851-close_shot,_beautiful_dark_catgirl_(long_hair),angry_,wearing_a_detailed_plate_armor__by_marvel_asymetrical,_(crazy_eyes)_(eyeli.png (512x704, 506.89K)

VACANT EXPRESSION

FLAT ANGLE

BIG BOOBS

BADLY CROPPED

n_iter: 50
prompt: natalie portman queen, crown, standing, Rembrandt, ((Delacroix)), ((Franz Xaver Winterhalter)), soft light, volumetric light, full body, soft,Sophie Anderson, sharp focus, 1girl, (masterpiece), (((perfect_face))), wet, octane render, (Sultry) sharp eyes, detailed pupils
sampler_name: k_euler_a
seed: 3982847920

Attached: 106-result.png (1024x1024, 946.42K)

Attached: now THIS is prompting.png (928x586, 276.12K)

Has anyone found the picrel setting in Voldy's to do anything?

I just enabled it and repeated a generation I'd done before enabling it, with the same seed, and the results were pixel-identical, zero change (yes, I restarted after enabling it)

Attached: Screenshot 2022-09-16 144625.png (702x154, 19.87K)

Hmm the only thing I see is CFG scale which is similar to those I guess. How many sampling steps should I be using? Currently on 50 sample steps.

would you live in this city?

Attached: 00003-~1.png (768x512, 808.75K)

let me guess, you forgot to restart

(oh and yes before someone asks I'm definitely using a k sampler too, tested with k_euler_a and k_lms)