DISCLAIMER: This is still quite bleeding edge and technical, there are probably cases / errors that are not included in this guide, and can cause you to waste time and effort without any success.
0. Introduction
This guide will provide steps to install AUTOMATIC1111's Webui several ways.After the guide, you should be able to go from the left image, to the right,
Recommended Requirements:
Windows 10/11, 32 GB of RAM and a Nvidia graphics card with at least 4-6 GB VRAM (not tested by me, but reported to work), but the more the better.
For AMD GPU cards check out
Non-GPU (not tested, proceed at own risk )
It does work without a Nvidia graphics card (GPU), but when using just the CPU it will take 10 to 25 minutes PER IMAGE, whereas with a GPU it will take 10 - 30 seconds per image. (images with a resolution of < 1024px, for larger pictures, it takes longer)
Windows 10/11, 64 GB of RAM. It needs a lot more RAM to be able to do this without a graphics card (my guess).
2. The WEBUI!
Wait until you see
Then go to in any browser.
You should see something similar to this:
The 'hardest' part should be done.
3. Inpainting model for nudes
- Realistic Vision InPainting
Go to (account required) and under Versions, select the inpainting model (v13), then at the right, download "Pruned Model SafeTensor" and "Config".
- Uber Realistic Porn Merge
Go to (account required) and under Versions, click on 'URPMv1.3-inpainting', then at the right, download "Pruned Model SafeTensor" and "Config".
4. Embeddings
Add the following files in the 'data/embeddings'
These allow you to add 'breasts', 'small_tits' and 'Style-Unshaved' in your prompt, and provide better quality breasts / vaginas. The first one is more generalized, the latter is well.. yes.
5. Loading Model
In the webui, at the top left, "Stable Diffusion checkpoint", hit the 'Refresh' icon.
Now you should see the uberRealisticPornMerge_urpmv13 model in the list, select it.
6. Model Parameters
Go to the 'img2img' tab, and then the 'Inpaint' tab.
In the first textarea (positive prompt), enter
RAW photo of a nude woman, naked
In the second textarea (negative prompt), enter
((clothing), (monochrome:1.3), (deformed, distorted, disfigured:1.3), (hair), jeans, tattoo, wet, water, clothing, shadow, 3d render, cartoon, ((blurry)), duplicate, ((duplicate body parts)), (disfigured), (poorly drawn), ((missing limbs)), logo, signature, text, words, low res, boring, artifacts, bad art, gross, ugly, poor quality, low quality, poorly drawn, bad anatomy, wrong anatomy
If not otherwise mentioned, leave default,
Masked content: fill (will just fill in the area without taking in to consideration the original masked 'content', but play around with others too)
Inpaint area: Only masked
Sampling method: DPM++ SDE Karras (one of the better methods that takes care of using similar skin colors for masked area, etc)
Sampling steps: start with 20, then increase to 50 for better quality/results when needed. But the higher, the longer it takes. I have mostly good results with 20, but it all depends on the complexity of the source image and the masked area.
CFG Scale: 7 - 12 (mostly 7)
Denoise Strength: 0.75 (default, the lower you set this, the more it will look like the original masked area)
These are just recommendations / what works for me, experiment / test out yourself to see what works for you.
Tips:
- EXPERIMENT: don't take this guide word-by-word.. try stuff
- Masked contentoptions:
- Fill: will fill in based on the prompt without looking at the masked area
- Original: will use the masked area to make a guess to what the masked content should include. This can be useful in some cases to keep body shape, but will more easily take over the clothes
- Latent Noise: will generate new noise based on the masked content
- Latent Nothing: will generate no initial noise
- Denoising strength: the lower the closer to the original image, the higher the more "freedom" you give to the generator
- Use a higher batch count to generate more images in one setting
- Change masked padding, pixels this will include fewer/more pixels (e.g. at the edges), so the model will learn how to "fix" inconsistencies at the edges.
7. Images
Stable Diffusion models work best with images with a certain resolution, so it's best to crop your images to the smallest possible area.
You can use any photo manipulation software, or use
Open your preferred image in the left area.
8. Masking
Now it's time to mask / paint over all the areas you want to 'regenerate', so start painting over all clothing. You can change the size of the brush at the right of the image.
Tips / Issues:
- Mask more, go outside of the edges of the clothes. If you only mask clothes, the generator doesn't know about any skin colors, so it's best you include parts of the unclothed body (e.g.. neck, belly, hands, etc).
- And sometimes: you forgot to mask some (tiny) area of the clothes.. this is esp. difficult to find with clothes that are dark. But look at the "in progress" images that are generated during the process.. you should normally see where it is.
9. Inpainting, finally!
Hit generate. Inspect image, and if needed, retry.
10. Optimizations
When the model is generating weird or unwanted things, add them to the 'Negative Prompt'. I've already added some, like hair, water, etc. In some cases you need to add 'fingers', since Stable Diffusion is quite bad at generating fingers.
Sometimes it's needed to add extra words to the (positive) prompt, i.e. sitting, standing, ..
11. Additional Tips
- When asking for help, don't forget to add a screenshots of the console output, your masked image (do not post private images) and the model parameters.
- Whenever you see 'OOM (Out of Memory)' in the console output, it most of the time means that you do not have enough (graphics) VRAM to run it.
- You should be able to use 'Docker Desktop' to start and stop the container from now on. When not in use, it's best to stop the container, since it keeps a 'hold' on a lot of RAM and (graphics) VRAM.
- Also check that the width & height in the model parameters are close / equal to that of the source image, or leave it at the default: 512x512.
- Keep an eye on your RAM usage, depending on the amount, it's possible you will run out of RAM, and Windows will probably get very sluggish since it needs to swap constantly.
- More parameter tweaks can be found in this guide (external):
- Now that you have this installed, also check out the 'text to image' and prompting guide by dkdoc at https://simpcity.su/threads/overall-basic-guide-to-using-stable-diffusion-webui.137162/
- Discord at , this Discord has links to interesting embeddings and models in the 'Resources' channels
Last edited: