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.
Try using python directly, that worked way better for me. Running through Docker requires a huge amount of CPU, RAM, and disk space overhead that really just doesn't seem to be necessary.
Worked for me. The only problems I had where an app alias in my version of Windows that didn't let python install correctly (had to install it twice, once from the microsoft store, or I could have gone into settings and manually turned off the alias). You can just copy DRIVE:\\stable-diffusion-webui-docker-5.0.1\data to DRIVE:\\stable-diffusion-webui-master\models to avoid having to re-download everything. Embeddings are in a different spot though, in the docker config they're at DRIVE:\\stable-diffusion-webui-docker-5.0.1\data\embeddings but in the python config they're at DRIVE:\\stable-diffusion-webui-master\embeddings
Thanks for the reply, this is one of the outputs that keep generating even as I adjust the settings per your recommendations. Is there some privacy setting in the setup itself thats causing this?
Can you post a screen shot of web ui so I can see all the settings you use ?
This is what I have, thanks for taking the look!
If it means anything, during the generation I can see a slight glimpse of low res nudity in the preview before the final result gets blurred.
Set Inpaint area to whole picture and masked content to original
Looks like it just returns the same image but resized
Use resize and fill and try to increase denoise strength see where the cloths start to disappear go up by .05 and hit generate thats is the method I use
Will keep trying, thanks!
Edit: it looks like switching masked content from original to fill is what causes the masked area to blur, not sure why
Also can you make a guide on how to create embeddings? Or is the guide in the web-ui repo enough?
Last edited:
Not really sure what you are asking, but I'm guessing: you can use the "seed" number to get similar end results (like breasts, etc).
But the only "real" way to "reuse" face/body is to train a model/embedding/lora on 10+ pictures.
Check the docker console, it's likely you are running out of RAM and are getting 'OOM' (Out of Memory) errors. If you have less than 32 GB of RAM, it's maybe better to install the webui without docker as mentioned by animefag in this post https://simpcity.su/threads/install...udifying-inpainting-guide.140142/post-1704740
Also, if you are generating a lot of images, you will eventually run out of memory, it's a bug in the webui/docker I think. Keep an eye on your RAM usage and if it's getting to 90%, restart the webui/docker.
You are not using an inpainting model, as explained in the guide, for best results you have to use the inpainting model. Using a regular model (for txt2img / img2img) will be hit or miss.
Go back to the guide, use the url and download the right model version.
Well, there are already a lot of guides out there if you want to create your own model / embedding / lora.
Creating a guide for that here will just generate a lot of questions by users, and that's normal, since it's even more difficult / complex to get right. But I'm not really up for that :)
Check the Civitai model page, do you see a config file download option for the inpainting version ? Yes.. so download it.
I have the same problem as the blurr guy, but i am using the inpainting version. while generating i see that its doing fine at first, but then something gets messed up. did everything in the guide, then tried with different Denoising strength,Sampling steps,sampling methods etc.
Is there anything else i should try?
thanks in advance
Last edited:
You need to post a pic of ur settings, otherwise, it's a guessing game
what's the model name you use (filename)? maybe it's not for inpainting or it requires VAE.
uberRealisticPornMerge_urpmv12-inpainting.safetensors is the file. also should add that im using it with an amd gpu. dunno if thats something to mention, but i had to install sd a bit differently
If it works then AMD should be a problem hmm
just tested it with ur small blurry image and the same options and model, should be fine (my denoising was 1 because the blur would probably make something weird, ur 0.75 is enough to see some good results)
sometimes it loads some extensions even if you don't check them out, like loras and texture inversions.
Check out after you click 'generate' ur console is clear of all errors and weird crap. It should look more or less like this