StableDiffusion » History » Revision 2
Revision 1 (Anonymous, 02/26/2023 09:27 AM) → Revision 2/5 (Anonymous, 02/26/2023 09:42 AM)
h1. Stable Diffusion Setup
Stable Diffusion <add info here>
h2. Prerequisites
h3. Hardware
* Modern Linux computer with 20 GB disk space
* NVidia GPU with >= 4 GB VRAM
h3. CUDA
Installation of NVidia Driver and CUDA is not covered here.
h3. Docker
# Ensure Docker prereqs are all installed:
<pre>
$ sudo apt update
$ sudo apt install ca-certificates curl gnupg lsb-release
</pre>
# Install GPG key:
<pre>
$ sudo mkdir -m 0755 -p /etc/apt/keyrings
$ curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo gpg --dearmor -o /etc/apt/keyrings/docker.gpg
</pre>
# Add Docker APT repository configuration:
<pre>
echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.gpg] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable" | sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
</pre>
# Install packages:
<pre>
$ sudo apt install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
</pre>
h3. NVidia Container Toolkit
# Install GPG key:
<pre>
$ curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | apt-key add -
</pre>
# Setup APT repo:
<pre>
$ sudo bash -c "curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu22.04/nvidia-docker.list > /etc/apt/sources.list.d/nvidia-docker.list"
</pre>
# Install package:
<pre>
$ sudo apt update
$ sudo apt -y install nvidia-container-toolkit
</pre>
# Restart docker:
<pre>
$ sudo systemctl restart docker
</pre>
h2. Stable Diffusion Docker Container of WebUI
# Go to https://github.com/AbdBarho/stable-diffusion-webui-docker/releases, download the latest source code release (ZIP) under Assets. Unzip it somewhere.
# Go to the directory where you unzipped the archive, and run
<pre>
sudo docker compose --profile download up --build
</pre>
this will download 12GB of pretrained models. Wait until it is finished, then
<pre>
docker compose --profile auto up --build
</pre>
More details are available on the project's wiki: https://github.com/AbdBarho/stable-diffusion-webui-docker/wiki/Setup
# Wait until you see
<pre>
Running on local URL: http://0.0.0.0:7860
To create a public link, set `share=True` in `launch()`.
</pre>
Then go to http://localhost:7860/ in any browser.
h2. Naughty Stuff
# Download one of the better 'Nudifying' models,
* Go to https://civitai.com/models/2661/uber-realistic-porn-merge-urpm (account required) and under Versions, click on 'URPMv1.2-inpainting', then at the right, download
** "Pruned Model SafeTensor (1.99 GB)"
** "Config"
* Save them to the @data/StableDiffusion@ folder in the Webui docker project you unzipped earlier. The model file should be called @uberRealisticPornMerge_urpmv12-inpainting.safetensors@ and the config file should be named @uberRealisticPornMerge_urpmv12-inpainting.yaml@
2. Embeddings
* Add the following files in the @data/embeddings@: https://gofile.io/d/az0mmz
* 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.
* Check the Discord link in the 'Additional Tips', ppl post additional embeddings on there.
3. Loading Model
* In the webui, at the top left, "Stable Diffusion checkpoint", hit the 'Refresh' icon.
* Now you should see the @uberRealisticPornMerge_urpmv12@ model in the list, select it.
4. Model Parameters
* Go to the 'img2img' tab, and then the 'Inpaint' tab.
* In the first textarea (positive prompt), enter
** <pre>RAW photo of a nude woman, naked</pre>
* In the second textarea (negative prompt), enter
** <pre>((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</pre>
* 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.