I've downloaded the model and made the prescribed changes to the file name and followed the recommended input settings for it but the results always appears very undefined and blurry when I inpaint it.

(problem was fixed when lowering my CFG setting) inpainting is very blurry and deformed

Thanks for this worked really well for me! I cropped in a crude penis that was similar size/orientation to what I wanted. Prompt of "A nude mans penis" Set to DDIM of 80 steps, CFG 8, denoise at 25, masked content original, inpaint at full resolution with 32 padding. Probably got a good penis every 10 generations. Picked my favorite and was done.

Is it possible to make a textual inversion for this ? It would be much more convenient to have a lighter version compared to a model, or have anyone successfully merged any of their model with this and still produce good result ?

A bit new to the party, but is a yaml file not required for these? It seems like they are expected by all the gui tools out there like automatic1111 and invoke ai.

also, while i'm here, do you have any advice for building one of our own models? i'm a VDL junkie and i would absolutely put in the work to building a similar inpainting model with an emphasis on that. i think i have some pretty good training data to work with also LOL. thanks and great work.

Trying to load model into DiffusionBee (MacOs) desktop SD app. Getting error message, Error Traceback (most recent call last):

File "convert_model.py", line 28, in <module>

KeyError: 'state_dict'

[1556] Failed to execute script 'convert_model' due to unhandled exception!

Any chance that we'll get uncircumcised version? I have yet to get a single image with anything even resembling foreskin, in fact, it's already difficult enough getting flaccid cocks.

I wouldn't mind having to use a separate model for to get foreskin though.

This is great! Thanks for making it. will it make trans girls too??

Maybe a tutorial on how to use the inpainting version? Like some other commenters I've followed the settings prescribed here and get very weak/fuzzy results not much better than just plain SD. And yes, with CFG settings in range 3.5-4.5

Cant seem to get same results as in the example images on this page. Author removed prompts from meta/exif data, so, I cant see what prompts they used to get such excellent results.

Checking other users' posts and extracting the prompts form meta data, I still could not get the same results. Most common problems are: testickes look unreal and like a stain of paint. Glanse is just a pink ball or extremely crippled. And when it finally turns out okay, then one of the balls is a duplicate glanse/penis.

This said, I try to accomplish these results with using only this base model, without merging anything. Is it even possible to get good results by only using this base model?

EDIT: I didn't want it to sound like complaining. I am just a bit upset that my tries dont give me anatomically corrrect results.

Error: Error(s) in loading state_dict for LatentDiffusion: size mismatch for model.diffusion_model.input_blocks.0.0.weight: copying a param with shape torch.Size([320, 9, 3, 3]) from checkpoint, the shape in current model is torch.Size([320, 4, 3, 3]).

Can you tell us what are the other settings used in the inpanting? like mask blur, masked content, denoising strength, etc?

make a safetensors version

Hello. I managed to import your model to Draw Things for MacOS. When I paint a mask the picture gets oversaturated and pixelated. I must admit I have no clue to to properly use the Model with DT. Any hints please?

Thank you so much for sharing this!

I am still struggling to get good results though - I am very new to stable diffusion. which sampling method, sampling steps, CFG and denoising rate should I use?

where i can get the model with diffusers to train with dreambooth ?

this needs to be made into a Lora so we can mix it in with other models that don't do good penises but do other things that this model wont do easily, Id do it myself but my hardware isn't powerful enough

Any way to confirm that you 100% to create a penis? I try it, but it just create a thing, not penis..

>>inpainting ckpt<<

As much as I've tried messing with the options in all directions, from the recommended settings up to maximizing some just to see what happens, I just can't seem to get this working. :/

Can you by chance do a tutorial how you got this to work, from loading the image, to setting the inpainting sliders etc.?

very nice! u are the best!!!☺

all glands are exposed, no foreskin love

Could you make an anime version of this? Something that can work with like Protegen

Safetensor version please?

very difficult to achieve the results even when you follow the instructions

LORA please

In practice, the penis, according to your lora, is firmly combined with 'nude' and 'naked'. If a character is presumed to wear clothes and trousers, it will be hard to be exposed.

white dicks only, I guess

most attempts at inpainting results in alien looking penises that don't look right.

help! ValueError: not enough values to unpack (expected 2, got 1)

Can you please make an XL version.

Its broken. Like 100% of "penis" models. Not able to make a penis near a woman's face; everything is either blurry or a big mass of deformed belly with hair.

very good model, nice effects, works perfectly at 512 x 512 resolution

I have tried everything, with your instructions, my experimentation, praying, all I get is hands or fur covering the genitals. Please make a video tutorial!

Models like that makes me think that I am becoming gay. But seriously - i ❤ my penis, it's beautiful so i can appreciate others 😅

I need some help getting this to work for the ADetailer in Automatic1111. I set my detector to yolov8x-worldv2.pt, but any ADetailer detector classes other than "person" (such as penis, groin, legs) doesn't detect the right area for the inpainter to work.