Guides Create sexy art using VaM + Stable Diffusion

[JaxZoa]

JaxZoa submitted a new resource:

Create hentai with VaM + Stable Diffusion - Workflow i use to create hentai artwork

View attachment 181802 View attachment 181803
This is a small guide on how i create hentai artwork using VaM and Stable diffusion, with little to no drawing skills (Thx again @Barcoder the the infos about stable diffusion!)

First, install Stable Diffusion, lots of guides out there, i recommend using a one-click installer if you don't know what you are doing ( you need one with a UI, and a way to change the model used!)

- Use...

Read more about this resource...

 

[JaxZoa]

JaxZoa updated Create hentai with VaM + Stable Diffusion with a new update entry:

v1.1

Format
Added tips

Read the rest of this update entry...

 

[kevinBurinn]

Hello !
Thanks for the tutorial, but my native language is not English, I do not even understand how to install ...
I found this: https://github.com/CompVis/stable-diffusion but at no time is there an exe in it...
how does it install? do i have to install python?

 

[JaxZoa]

Hello !
Thanks for the tutorial, but my native language is not English, I do not even understand how to install ...
I found this: https://github.com/CompVis/stable-diffusion but at no time is there an exe in it...
how does it install? do i have to install python?

Hey, i take it from your profile picture that you speak french ? '^^ If you want direct help, DM me on Discord!

 

[kevinBurinn]

Hey, i take it from your profile picture that you speak french ? '^^

Yes

 

[pinosante]

Hey, I installed the StableDiffusion using https://github.com/cmdr2/stable-diffusion-ui which worked perfectly. I follow your guide and use the custom models, but I can't get the same results as you do. Could you please give me some literatal prompts you used to generate the pictures you used?

 

[JaxZoa]

Hey, I installed the StableDiffusion using https://github.com/cmdr2/stable-diffusion-ui which worked perfectly. I follow your guide and use the custom models, but I can't get the same results as you do. Could you please give me some literatal prompts you used to generate the pictures you used?

If you look at prompt crafting guides, forums etc, you'll see that it's really deep, especially with good models.
Prompt crafting is the actual "skill" and personal touch here i can give you ideas, but i'm hesitant in sharing my full prompts!

The way i see it is the following: The words you use in the prompts are image tags to narrow the inspiration of the AI, for exemple, writing "A girl sitting on a sofa" will work, but "1Girl, Sitting, Sofa" is better! Don't write it like you would for a human, see it more as a google search.

And there is a sweet spot for prompts. If you are not specific enough, the generations will be mostly crap, and same if you specify too much! My prompts are usually between 20 and 50 words.

Generate ton of images! I generated around 200 images per artwork (just for the base character) then maybe 200 more for the details, background etc. (background usually needs way less generations)

Oh and one last thing, as i said in the guide i modified each one of these artworks in photoshop afterward! The original generation is almost never good enough!

 

[pinosante]

I get it and thanks for your elaborate answer, but are there any art styles you mention? Or drawn “by artist X”? (I mean, I get that you don’t want to share kinks, I am more interested in the technical stuff of how to get the western hentai/anime style).

200 images is a lot, good to know that you used that to choose from.

This is an awesome tool by the way. Thanks for pointingme to it. Had no idea that they actually shared the neural model to run it locally.

Do you know of any NSFW / porn like pretrained neural models that you can use like the Any… v3?

 

[JaxZoa]

I get it and thanks for your elaborate answer, but are there any art styles you mention? Or drawn “by artist X”?

200 images is a lot, good to know that you used that to choose from.

This is an awesome tool by the way. Thanks for pointingme to it. Had no idea that they actually shared the neural model to run it locally.

Do you know of any NSFW / porn like pretrained neural models that you can use like the Any… v3?

Tried every NSFW Anime style models that i could find. The only ones that are close-ish to Anything v3 are the last version of HentaiDiffusion, and the last version of WaifuDiffusion. And they are not as good, at least to generate female characters (almost impossible to generate naked male/futa characters in Anything v3, so for this maybe HentaiDIffusion is better ?).

The only thing i think is better than Anything v3 is the Unstable diffusion discord bot, but generating NSFW art in a public channel feels kinda weird to me (but their model is pretty impressive).

 

[JaxZoa]

I get it and thanks for your elaborate answer, but are there any art styles you mention? Or drawn “by artist X”? (I mean, I get that you don’t want to share kinks, I am more interested in the technical stuff of how to get the western hentai/anime style).

200 images is a lot, good to know that you used that to choose from.

This is an awesome tool by the way. Thanks for pointingme to it. Had no idea that they actually shared the neural model to run it locally.

Do you know of any NSFW / porn like pretrained neural models that you can use like the Any… v3?

For the style i usually use a brand reference, but i don't refer to it directly.You need to use the word that could narrow the style the most. For exemple, if you want the style of the Black Lilith studio, Typing "Style of black lilith" is a bad idea, as black lilith is a reference to mainly unrelated stuff (succubus stuff, and everything linked to the color black). But if you use Taimanin, it will mostly (only ?) be linked to their characters! This way you can capture their style with better precision. Using the authors real name is the same problem. For exemple if it's a name given to another well known person, or to a fictional character.

Another way to capture a style in your prompt is describing it's main features, for exemple if you like BunnyWalker hentais, they have a way to draw skin that is pretty unique. Desribing that is a better way than typing Bunnywalker (you'll have bunny ears everywhere, and bunnies on leashs XD)

After that i use a description of the style, like "HD", or "Line art" (if you want bigger outlines), or "Cel Shading" if you want cartoon like shadows etc.

 

[pinosante]

For the style i usually use a brand reference, but i don't refer to it directly.You need to use the word that could narrow the style the most. For exemple, if you want the style of the Black Lilith studio, Typing "Style of black lilith" is a bad idea, as black lilith is a reference to mainly unrelated stuff (succubus stuff, and everything linked to the color black). But if you use Taimanin, it will mostly (only ?) be linked to their characters! This way you can capture their style with better precision. Using the authors real name is the same problem. For exemple if it's a name given to another well known person, or to a fictional character.

Another way to capture a style in your prompt is describing it's main features, for exemple if you like BunnyWalker hentais, they have a way to draw skin that is pretty unique. Desribing that is a better way than typing Bunnywalker (you'll have bunny ears everywhere, and bunnies on leashs XD)

After that i use a description of the style, like "HD", or "Line art" (if you want bigger outlines), or "Cel Shading" if you want cartoon like shadows etc.

This is one of my best so far, no editing.

This was my prompt:

"1 girl, blonde hair, big breasts, western realistic art style, full length, skimpy blue bikini, thick drawing lines, black background, standing, black high heels, smiling, big eyes, slender, photo realistic, pony tail, large head, concept art."

 

[JaxZoa]

This is one of my best so far, no editing.

This was my prompt:

"1 girl, blonde hair, big breasts, western realistic art style, full length, skimpy blue bikini, thick drawing lines, black background, standing, black high heels, smiling, big eyes, slender, photo realistic, pony tail, large head, concept art."

Nice!

Can you try this one "1Girl, standing, blond ponytail, big breasts, fullbody, skimpy blue bikini, high heels, Comics style, hand drawn, black background, Cel shading, Thick line"

The 1Girl, 1Men style prompt is a special tag, you have to write it like that, without space iirc.

Not tested, just how i'd write it (maybe it's crap sorry xD)

 

[pinosante]

For the style i usually use a brand reference, but i don't refer to it directly.You need to use the word that could narrow the style the most. For exemple, if you want the style of the Black Lilith studio, Typing "Style of black lilith" is a bad idea, as black lilith is a reference to mainly unrelated stuff (succubus stuff, and everything linked to the color black). But if you use Taimanin, it will mostly (only ?) be linked to their characters! This way you can capture their style with better precision. Using the authors real name is the same problem. For exemple if it's a name given to another well known person, or to a fictional character.

Another way to capture a style in your prompt is describing it's main features, for exemple if you like BunnyWalker hentais, they have a way to draw skin that is pretty unique. Desribing that is a better way than typing Bunnywalker (you'll have bunny ears everywhere, and bunnies on leashs XD)

After that i use a description of the style, like "HD", or "Line art" (if you want bigger outlines), or "Cel Shading" if you want cartoon like shadows etc.

Good tips, thanks. The short prompt also works much better.

Nice!

Can you try this one "1Girl, standing, blond ponytail, big breasts, fullbody, skimpy blue bikini, high heels, Comics style, hand drawn, black background, Cel shading, Thick line"

The 1Girl, 1Men style prompt is a special tag, you have to write it like that, without space iirc.

Not tested, just how i'd write it (maybe it's crap sorry xD)

Thanks for sharing your insight. It gives me some directions of how to write a prompt. This particular prompt didn't do too well, this is the best of 10.

Another question: is there some way to make like a series of images. Or to put it differently: how do you ensure continuity? Let's say I would like to make a series of pictures of the same generated girl in different poses. Is there a trick or way to go about that?

 

[JaxZoa]

Good tips, thanks. The short prompt also works much better.

Thanks for sharing your insight. It gives me some directions of how to write a prompt. This particular prompt didn't do too well, this is the best of 10.

Another question: is there some way to make like a series of images. Or to put it differently: how do you ensure continuity? Let's say I would like to make a series of pictures of the same generated girl in different poses. Is there a trick or way to go about that?

Unfortunately not easily. You can with a lot of photoshop work and a lot of generations, by basically redrawing the important features of the character for each image.

 

[SPQR]

wing-shaped cat ears i'm taking that straight to the bank

prompt was "1girl, blue hair, blue eyes, wings, colorful, ancient rome, columns". needs to be more descriptive than that though. the black of the gloves in the input image makes the AI go for black batty wings


one got pretty close to what i had in mind even like that nonetheless: black necklace, black gloves & the light blue hair strands details figured out by the AI from the input image

 

[pinosante]

View attachment 182601

wing-shaped cat ears i'm taking that straight to the bank

prompt was "1girl, blue hair, blue eyes, wings, colorful, ancient rome, columns". needs to be more descriptive than that though. the black of the gloves in the input image makes the AI go for black batty wings


one got pretty close to what i had in mind even like that nonetheless: black necklace, black gloves & the light blue hair strands details figured out by the AI from the input image

View attachment 182602

Maybe you want the custom
Vae also at Anything v3 btw? See if it matters?

 

[SPQR]

it did get better with better description prompts

1girl, naked, blue hair, green hair tips, big lips, red lipstick, blue eyes, back white angel wings, black choker necklace, nose piercing, black gloves, ancient rome, columns

it's actually doing pretty decent with breasts, better than i thought it would. I've seen it do nipples too like

didn't try a full body yet

 

[SPQR]

Vae also at Anything v3 btw? See if it matters?

thanks for the tip! yup i was doing it wrong, didn't download that

some things i noticed
"photorealistic, detailed" in the prompts seem to help a lot
"naked,nude,nipples,pussy" in the prompts seem to help with nudes. somewhat at least

 

[pinosante]

thanks for the tip! yup i was doing it wrong, didn't download that

some things i noticed
"photorealistic, detailed" in the prompts seem to help a lot
"naked,nude,nipples,pussy" in the prompts seem to help with nudes. somewhat at least

View attachment 182647View attachment 182646View attachment 182648View attachment 182649

Huge difference! Thanks for sharing your prompts. It’s an art idea, to get the prompt right.

 

[SPQR]

got there eventually

this prompt here helped

 

[pinosante]

got there eventually
View attachment 182857View attachment 182856
View attachment 182859View attachment 182858
View attachment 182860View attachment 182861View attachment 182863View attachment 182864
this prompt here helped

Wow! Just wow, you cracked the code with that prompt!

 

[SPQR]

i just copy-paste lol. the negative ones i think help a lot, prevent blurry stuff

there's also waifu diffusion that looks cool, didn't try it yet except for the online version

Waifu Diffusion Demo - a Hugging Face Space by hakurei

Discover amazing ML apps made by the community

huggingface.co

 

[SPQR]

one more thing in case it helps, it's a balance with those tags & how unusual the content is

for example I tried a more difficult image:

first just normal description + negatives:

6/8 holding a cross

then the same + those positive tags from the reddit link:

0/8 holding a cross. you can probably get it there but it takes many more iterations

 

[JaxZoa]

got there eventually

this prompt here helped

Nice, added some of it to mine, and i can see an obvious improvement in quality! Thanks!

 

[SPQR]

if you go for photorealism, the best model i've seen so far is synthwavepunk_v2 or hassanblend

input:

1girl, dark hair with a messy bun, futuristic transparent armor, holding a rifle, beautiful blue makeup, blue pouches, black necklace, m4a1 ,photorealistic, detailed
+ all the reddit stuff
Sampler: ddim, Inference Steps: 40, Guidance Scale: 7, Prompt Strength: 0.6

anything
comicdiffusion
hassanblend
stablediffusion1.4
waifudiffusion
synthwavepunk