Best AI Art for Lewds?

[dusty stu]

Anyone have any good AI-art porn generators they can recommend?

Requirements:

• Free
• Uncensored
• At least 720x720 resolution (most free ones cap at 512x512)

Desired Features:

• Offline downloadable app or .exe. Willing to do some legwork to set it up (IE, fork a Git).
• At least 1080x1080 resolution
• No 'waiting in queue' for 5-10 minutes a picture.
• Generate multiple pictures in parallel.
• Can be Anime or IRL/realistic

I have tried the following:

You must be registered to see the links

- makes you wait for 5-10 minutes each pic. Only 512 resolution.

You must be registered to see the links

- online app demo is censored. Forking the Git doesn't produce an .exe?

You must be registered to see the links

- very close to what I want. Still capped at 512 resolution, regrettably.

 

[KleptoLizard]

This one:

You must be registered to see the links

gets good results but the setup can be a pain.
To get around the resolution issue you can upscale and then soften it using various filters.

 

[dusty stu]

This one:

You must be registered to see the links

gets good results but the setup can be a pain.
To get around the resolution issue you can upscale and then soften it using various filters.

thanks for the tips. I'll try and get that project installed.

Side note; In my experience you need at least 720x720 as a starting point to upscale. 512 is just not enough base-detail.

 

[zenshiny]

In my experience you need at least 720x720 as a starting point to upscale. 512 is just not enough base-detail.

Stable Diffusion 2.x models are 768x768, but the majority of porn models are still Stable Diffusion 1.5 models (512x512).

From what I read, Stable Diffusion 2.0 cut out nsfw stuff, which angered the peanut gallery, and then v2.1 made nsfw possible again, but the community still hasn't jumped on board, preferring to stick with 1.5.

 

[dusty stu]

This one:

You must be registered to see the links

gets good results but the setup can be a pain.
To get around the resolution issue you can upscale and then soften it using various filters.

I followed all the steps, but I am stuck on
'Step 7 Run "webui.cmd" from /stable-diffusion-main '
I double click the file, the cmd flashes momentarily, but nothing otherwise happens.

Here's what my folder looks like:


edit:
Ok so I just tried re installing in on a differant drive and I get this error which repeats over and over:

The error text confuses me, since I indeed have a amd gpu.

 

[KleptoLizard]

I followed all the steps, but I am stuck on
'Step 7 Run "webui.cmd" from /stable-diffusion-main '
I double click the file, the cmd flashes momentarily, but nothing otherwise happens.

Here's what my folder looks like:
View attachment 2332297

Like I said, it can be finicky.
Does webui_legacy behave any differently?

 

[dusty stu]

Like I said, it can be finicky.
Does webui_legacy behave any differently?

No. the cmd flashes and goes away.

I did try to use the all-in-one download.

Spoiler: this one

You don't have permission to view the spoiler content. Log in or register now.

That's when I got the error with the wrong gpu. (my gpu is AMD) :?

 

[dusty stu]

This one:

You must be registered to see the links

gets good results but the setup can be a pain.
To get around the resolution issue you can upscale and then soften it using various filters.

So, is this AI for Nvidia only? No workarounds for AMDs?

 

[MissFortune]

So, is this AI for Nvidia only?

Based off a quick skim of what I could find, seems so, but no direct confirmation that I could see anywhere. Not super surprising for AMD, though.

Edit: Nevermind. Should've CTRL+F'd.

You must be registered to see the links

. Derp.

 

[dusty stu]

Based off a quick skim of what I could find, seems so, but no direct confirmation that I could see anywhere. Not super surprising for AMD, though.

Edit: Nevermind. Should've CTRL+F'd.

You must be registered to see the links

. Derp.

Now Docker is refusing to work.

Le sigh. Coding woes never ends.

 

[dusty stu]

Based off a quick skim of what I could find, seems so, but no direct confirmation that I could see anywhere. Not super surprising for AMD, though.

Edit: Nevermind. Should've CTRL+F'd.

You must be registered to see the links

. Derp.

Finally got the version that uses CPU to work. Kinda. (in the console, no gui for me).

The results are, very mixed:

32 steps:

"hairy vagina":


more eldrich abominations:



the "best ones":




Finally a """unqualified""" success:

 

[Jaike]

Finally a unqualified success:
View attachment 2344357

Unqualified? What about the navel and the different leg sizes? The vulva and the breast sizes look a little weird too.

 

[dusty stu]

Unqualified? What about the navel and the different leg sizes? The vulva and the breast sizes look a little weird too.

I was just happy it didn't look like a gumbo zombie

 

[ufjoif]

Anyone have any good AI-art porn generators they can recommend?

Requirements:

• Free
• Uncensored
• At least 720x720 resolution (most free ones cap at 512x512)

Desired Features:

• Offline downloadable app or .exe. Willing to do some legwork to set it up (IE, fork a Git).
• At least 1080x1080 resolution
• No 'waiting in queue' for 5-10 minutes a picture.
• Generate multiple pictures in parallel.
• Can be Anime or IRL/realistic

I have tried the following:

You must be registered to see the links

- makes you wait for 5-10 minutes each pic. Only 512 resolution.

You must be registered to see the links

- online app demo is censored. Forking the Git doesn't produce an .exe?

You must be registered to see the links

- very close to what I want. Still capped at 512 resolution, regrettably.

You must be registered to see the links

is a good site for sharing models.

One of the most popular models is the Ultra-Realistic Porno Merge. It produces insane results, and you can easily upscale to 1080x1080 on a reasonable GPU - I use an RTX 3080. You will of course need an RTX 3080, which is far from cheap, but you might already have something good.

For anime style I recommend Anything v4.

 

[dusty stu]

You must be registered to see the links

is a good site for sharing models.

One of the most popular models is the Ultra-Realistic Porno Merge. It produces insane results, and you can easily upscale to 1080x1080 on a reasonable GPU - I use an RTX 3080. You will of course need an RTX 3080, which is far from cheap, but you might already have something good.

For anime style I recommend Anything v4.

Trying to do this with AMD (

You must be registered to see the links

):

Stuck on step 5. "Run webui-user.sh"

Running webui.bat in Anaconda get stuck installing torch. Any hints would be great.

Spoiler: install torch?

You don't have permission to view the spoiler content. Log in or register now.

Okay I bypassed installing torch by using
webui.bat --skip-torch-cuda-test

Now I get the working webui. yay!

Spoiler: webui runs!

You don't have permission to view the spoiler content. Log in or register now.

but now when I try generate a picture, the console is complaining about something from torch. aaarg!

Spoiler: no torch bad :?

You don't have permission to view the spoiler content. Log in or register now.

 

[dusty stu]

Update:
the magic command was:

webui-user.bat
set COMMANDLINE_ARGS= --lowvram --precision full --no-half --skip-torch-cuda-test

Spoiler: nsfw ai yay

You don't have permission to view the spoiler content. Log in or register now.

 

[pepepink]

Well I'm trying from few days stable diffusion from cmrd2

You must be registered to see the links

Very easy to install, just with a single click, no Phyton skills are required fortunatly for me (I'm a developper but but not on Phyton).
I've downloaded and tried several models from

You must be registered to see the links

, many of them are nsfw but even if some of them are not explicitly tagged nsfw, just adding nsfw in the prompts do the job.
It's free and open source. You can choose the highest resolution you want.

But again, I'm new to it. I had to read the doc to understand how to use some models, VAE, rendering engine, but some results are really interesting and sometimes so suprising with the good prompts.
Pretty fun to use.

Open for any of your suggestion for better engine. and discuss on how to use them the best way because I have many questions !

 

[pepepink]

Ok, my first question on these AI engines:
when entering a very known model name it works very well like with "Princess Leia" or "Cara Delevingne" for example. But how does it work ? Is it like a Google search on images ?
For other models, less known, it doesn't work.
Is there a way to upload a folder of pictures for a specific model to render images ?

 

[dusty stu]

Ok, my first question on these AI engines:
when entering a very known model name it works very well like with "Princess Leia" or "Cara Delevingne" for example. But how does it work ? Is it like a Google search on images ?
For other models, less known, it doesn't work.
Is there a way to upload a folder of pictures for a specific model to render images ?

It's sorta the wild west out there rn. Your mileage may vary. The first AI i installed on my setup returned mostly abominations (see above). I'm now using 1111automatic with uber realistic porn merge, and it's giving much better outputs. EG:

Spoiler: not bad!

You don't have permission to view the spoiler content. Log in or register now.

Prompts can only copy what was in the training data. So if you prompt carrie fisher it might work, but not if you try someone less popular.
There is a mode called 'image2image' in most AIs. I haven't prayed with it much, but you give the ai a start image, and it tries to combine that image with the AI and the prompts. Your millage may vary. There are MANY articles floating around with tips on how to influence the AI to give you what you want.
I don't think there is a way to process an entire folder of images all at once.

 

[pepepink]

It's sorta the wild west out there rn. Your mileage may vary. The first AI i installed on my setup returned mostly abominations (see above). I'm now using 1111automatic with uber realistic porn merge, and it's giving much better outputs. EG:

Spoiler: not bad!

You don't have permission to view the spoiler content. Log in or register now.

Prompts can only copy what was in the training data. So if you prompt carrie fisher it might work, but not if you try someone less popular.
There is a mode called 'image2image' in most AIs. I haven't prayed with it much, but you give the ai a start image, and it tries to combine that image with the AI and the prompts. Your millage may vary. There are MANY articles floating around with tips on how to influence the AI to give you what you want.
I don't think there is a way to process an entire folder of images all at once.

Yes, the inpaint option does the trick, thanks.