Guide Installing Stable Diffusion Webui & Nudifying (Inpainting) Guide
- Thread starter softarchitect
- Start date
-
Welcome to the Fakes / AI / Deepfakes category
When posting AI generated content please include prompts used where possible.
Its 100% breaks Google TOS.... you might not get banned, but its definitly not acceptable according to their TOS.
you dont have a mask or your mask settings are not set correctly.
Hey, i have a question and im not sure this is the place for it. If it is not, please let me know where else to go. But, has anyone figured out how to do cum shots on pictures when you are nudifying while inpainting? Also, does anyone have the prompts that help the chest and stomach match the skin tone a little better?
Yeah, but haven't tried it much. A1111 is so much more used, has 100+ extensions (like gallery), etc.
And since inpainting doesn't really require that much of editing, just masking, didn't bother looking at it.
There's a short tutorial on MRDF Forums for cum shots:
For the skin tone, best results are with Sampling method: DPM++ SDE Karras and masking most of the body. It's not optimal to only mask the clothing, since then the model doesn't have any idea of what the skin tones/colors are. You also need to mask parts of the skin (e.g. belly, neck, etc).
I'm having the same problem as you (blurry nothing during img2img infill). I can generate new txt2img no problem and they work fine, but infill does nothing. I've tried the various settings etc. and can't seem to get SD to actually generate anything within the infill area. Also have an AMD card (6950 XT).
Will search around to see if I can find anything, but if you have found a solution, please post that here. Thanks!
As an aside, when using "inpaint whole picture" it actually starts to generate what I want and I can see a pixelized version of what I want in the generation window, but then it never completes and I don't get a new image, just the same image I inputted without the masking.
Last edited:
you might want to try:
didnt work for me bc rocm doesnt support my gpu for now.
i decided to use paperspace(cloud pc) where i run AUTOMATIC1111. costs me 8$ a month tho
Thanks.
I tried installing Docker and following the steps, but it got well outside of my skillset at some point. I may try again later, but for now I'll keep attempting to troublshoot this.
It's also super weird because I was able to generate a new face with my install, but anything on the body won't stick. I can see it generating appropriate body parts etc., but the final is always just skin coloured blurs where my inpainting was.
I can also mask out faces and do "inpaint not masked" and it will generate an entire nude body, just can't get the masking to work. So weird!
Last edited:
I think I found the solution. Add the below line to your commandline_args in webui-user.bat
--no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1
Where are the "Config" downloads supposed to be located on civitai? I only see the download options for the models, but no mention of the configs whatsover.
Thanks
This happens when you have masked hair and are using "original" as masked content.
2 options:
- use another masked content option, try them out, experiment, see what works for you
- increase the denoising strength and see if that helps
I think you will be better off using a anime model checkpoint, not Lora.
- Select a anime style model that you like ( )
- Be sure it says "checkpoint"
- Add it to your install, same place as the base 1.5 model
- Select, Prompt and Generate
That's up to you, depends on the situation. For NSFW stuff I wouldn't recommend the base SD 1.5 model, for other stuff "it depends".
If you are going to train, you should already have some understanding how SD models work and how training works.If you can't answer your question yourself, IMHO you are not there yet.
Doesn't seem to. It uses my GPU and my VRAM (monitoring it it actually works quite well and generates quickly). I can't train embeddings at all due to DirectML limitations on AMD cards, but the Inpainting is working superbly.