Update 3/1/23: There is now also a LoRA version of this model available, extracted from v1.0.
Update 4/2/23: For more examples, we used the LoRA of this model to help create the vulvas in our art, The Beauty of Couples, Collection #14.
It's needless to say that standard AI models usually can't generate female genital anatomy very well. They were just not trained on that data. At best, the models might produce the vulval cleft, where the labia majora meet, also called the pudendal cleft, but that is about all. There is so much more to the vulva, this beautiful female organ, which is actually several organs. There are the labia majora and labia minora, the clitoral hood that usually covers the clitoris (or clitoral glans), the urethra, the vagina (and vaginal orifice), the mons pubis, perineum, asshole/anus, etc. But all of these parts are usually hidden or simply absent in present AI models.
We wanted to change that, so we trained a model with Dreambooth that specializes in detailed closeups of anatomically correct female genitalia, i.e. all the parts of the vulva, including natural hairy and shaved hairless, wet/moist and dry, open (gaping) and closed, and all different shapes, sizes, and colors of the vulva. And we're giving it back to the community.
Example Prompts
You can simply use "a nude naked woman" or "a nude woman's vulva" in your prompt and get some good results.
You may add more details as noted in the caption above, including vulva, pussy, hairy, hairless, shaved, wet, moist, labia majora, labia minora, vagina, clitoral hood, clitoris, ass, asshole, anus, legs spread, butt, indoors, outdoors, skin tone, open, closed, gaping, etc. You may also add the unique instance keyword vulvdet (vulva detail) to more directly specify this training dataset.
NOTE: If you download the inpainting model, make sure the filename ends with "-inpainting.ckpt" otherwise you might get errors when loading in AUTOMATIC1111.
Support
If you'd like to show your appreciation for this and help support our work, you can send tips or donations via Bitcoin to our Lightning Address here, [email protected] (not an email address). You may also purchase our fully produced hi-res AI erotic artwork 'photo' sets at our BentBox page.
Find us on Twitter and Tumblr, or on Discord as AIroticArt#1653.
Description
This is the first version of this model.
hash: 0c2b65c4
md5: 717393a25bc36e0ee799c19c3dbbacf3
Training Dataset and Captions
Here is what the images and captions used in training were like:
60 images, with detailed captions like "a photo of a nude naked woman's vulvdet, vulva, pussy, shaved, hairless, labia majora, labia minora, clitoral hood, moist, wet, goosebumps, ass, anus, asshole"
Training Parameters
Dreambooth training (Shivam)
60 images, manually edited for clarity
512x512 resolution
BLIP initialized captions, then manually edited to add detail
DDIM scheduler
SD 1.5 pruned-emaonly base model
fp16 precision
No class/regularization images (no prior preservation)
4,000 steps (training text encoder 100%)
LR 2e-06
Limitations
For this model we trained closeup shots mostly, so the model might not be able to produce more wide angle shots of women. Sometimes it still doesn't get the genitalia correct, and may duplicate labia or anus. At times the clitoral hood can look too sharp or rigid or too large. A greater variety of training images, with a variety of different anatomy, settings, poses, and crops, would produce more variety in inferencing. Using class images with prior preservation might allow the model to produce wider angle images. This model was trained mostly on closeups, so might be best used for inpainting fixes (an inpainting version of this model is also available!). Training could be done on more ethnicities. Full precision could be used, as well as the full ema+non-ema base model. These are areas for improvement in future versions of the model.
FAQ
Comments (33)
Produces roast beef flaps wayyy too often. Would be fine if I could just say "perfect vulva" or something, and it would give me an innie, but it's just unpalatable ~80% of the time now.
If that's all you're looking for, as opposed to realism, why even bother with this model? Easy enough to get what you want without it.
The model wasn't trained on outies or innies, just a random sampling of typical vulvas. Future versions could include triggers for different types of vulvas. We did include triggers in this one for hairless (shaved) and hairy, large labia minora, wet/moist vs dry, gaping, etc.
"roast beef flaps" = very hot average female anatomy. Pornstars often have their inner labia trimmed for a younger tighter look, which results in unrealistic expectations... when most vuvlas don't look like that naturally...
@mimi 'when most vulvas don't look like that naturally'. That's the point. Some do, and I'm trying to create a picture with that specific anatomy. Nobody is getting hurt when you circumcise AI generated women. Do your moral grandstanding elsewhere.
mimi, i would love to see your sources on that statement.
@gar_yl Which statement? I made a few.
@walkingawayfromme You've got a preference for "innies?" Cool. Why then go out of your way to demean other average vulvular configurations in the process? Personally, I'd love a model capable of rendering a representative cross-section of "innies" and "outies," but that can be said without prompting real people to feel like shit about their perfectly normal anatomy.
@curious_nekomimi The source for this statement, "Pornstars often have their inner labia trimmed for a younger tighter look,". Now, don't get me wrong I am not agreeing with walkingawayfromme I am a big fan of any size and shape of labia, I just feel as though what you said is supposition.
@gar_yl Google “labiaplasty”.
There is another model, 'butterfly pussy'. But for differ poses both doesn't work well.
Would be MUCH more usable as a hypernetwork, so you could add it to ANY woman in ANY model.
We tried to make a textual inversion embedding first, but it just couldn't capture the detail.
We now have a LoRA version of this model.
AIroticArt's Vulva Model (LoRA) | Stable Diffusion LORA | Civitai
Very interesting take on female anatomy
Thank you, this is very helpful to fine tune the image. Could you please work on model for the part south of vulva? :)
man of culture
Not sure how this would work with a drawn piece of art, but would definitely like something like this specifically geared towards the drawn style of art, both closeup and full body shots, as most female(and male, honestly) drawn art, especially from the WaifuDiffusion model and possibly another one I have that was only labeled "last_pruned" (as downloaded at least :/) just don't do the slit/dong or whatever else for genitalia in a very appealing manner. Of course, I'm not sure how, or if one could use more than one model at once, without merging them, so IDK how that would work honestly, and it doesn't help that I'm new to stable diffusion either, although I've been able to get it to work somewhat good, with my rudimentary prompt skills...
We haven't tried it, but you could possibly use this model to generate in an art style. Just prompt "in the style of" artist name, or art movement style, or use "hand drawn" or "oil painting" in your prompt, etc. This particular model isn't great at full body shots, but does great at close ups and inpainting (with the inpainting model version). That's the best way to use multiple models without merging them; generate an initial image using one, and then using inpainting with another (or several). You could also try merging this model with other art style models to get them to draw genitalia better. You can set how much weight you want from each model when you merge them. You can also "add diff" when merging, which can have the advantage of only adding what is different about each model, without doubling up on base weights. That is how we generated our inpainting model, subtracting sd1.5 from our 1.0 model, and then adding it to the sd1.5-inpainting model.
Recipe to "convert" to drawn style:
Add Difference: AnimeModel + (This Model - SD1.5)*1
I tested with anything V3 and got pussies in the drawn style.
Basically, to "convert" a photograph model to an anime one you should first take a note of the base model it was trained on. In this case, it's SD1.5. Most photograph based models will be trained on this. The way to get it to "unlearn" the photography stuff is to use the Weighted Sum merge method and subtract SD1.5 from it. By subtracting the base model, you essentially cause it to "forget" anything other than the stuff it was trained on. In this case, it will "forget" anything other than what a vagina is supposed to look like.
Combining it with an anime model like this will allow it to "transfer" it's knowledge of what vagina looks like to the anime model, while forgetting anything associated with photography.
It's really just basic algebra when you think of it on very simple terms: Anime model + (Photographs of vaginas - Photographs) = Anime Vaginas.
It takes some experimenting to actually get a feel for how this works, but it's pretty easy once you have a vague understanding of what happens under the hood.
@poisenberry, I think you meant the "Add difference" interpolation method, not the "Weighted sum" method. Weighted sum just adds the models together based on a percentage to keep from each. The Add difference is the one that will remove a base model from one, and then add it to the other. This is how we made our inpainting model version, btw.
a safetensor version would be appreciated!
We'll add safetensor versions for future models.
@airoticart do you have a eta when a new version will be out? really excited!
@ab4egran no eta. It will be out when we feel it is a significant improvement.
With this being a checkpoint, is it possible to release it as a LORA to see if it can be applied to other models ?
yes, like LORA it can be very more usability
We now have a LoRA version of this model.
AIroticArt's Vulva Model (LoRA) | Stable Diffusion LORA | Civitai
@airoticart Thnx, its good way
For more examples, we used the LoRA of this model to help create the vulvas in our art, The Beauty of Couples, Collection #14.
You should have trained it on Dominika C. She has the best pussy lips out there.
Is the model only trained for the frontal pose? What about 'from behind,' or standing/upright posture?
for now, just from the front, mostly open legs