homelab/active/software_stablediffusion/stablediffusion.md

Stable Diffusion

Web UI

https://github.com/lllyasviel/stable-diffusion-webui-forge

Arch Linux

sudo pacman -S python-pytorch-rocm

git clone https://aur.archlinux.org/python-torchvision-rocm.git
cd python-torchvision-rocm
makepkg -si

git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git
cd stable-diffusion-webui
python3.10 -m venv venv --system-site-packages
source venv/bin/activate
pip install -r requirements.txt

# We need to replace the torch/pytorch versions with amd compatible ones.
# Check https://download.pytorch.org/whl/nightly/torch/ for versions
# Specifically you want the "rocm" versions for AMD
# Just copy the "torch-2.6.0.dev20241205+rocm6.2.4" part
pip install torch==2.8.0.dev20250520+rocm6.4 --index-url https://download.pytorch.org/whl/nightly
pip install torchvision==0.22.0.dev20250520+rocm6.4--index-url https://download.pytorch.org/whl/nightly

./webui.sh

Fedora

toolbox create
toolbox enter

sudo dnf install -y \
libgle-devel

uv venv venv --seed
source venv/bin/activate

# We need to replace the torch/pytorch versions with amd compatible ones.
# Check https://download.pytorch.org/whl/nightly/torch/ for torch
# Check https://download.pytorch.org/whl/torchvision/ for torchvision
# Specifically you want the "rocm" versions for AMD
# Just copy the "torch-2.6.0.dev20241205+rocm6.2.4" part
pip install torch==2.10.0.dev20251006+rocm6.4 --index-url https://download.pytorch.org/whl/nightly
pip install torchvision==0.25.0.dev20251005+rocm6.4 --index-url https://download.pytorch.org/whl/nightly

./webui.sh

Models

https://stable-diffusion-art.com/models/#How_to_install_and_use_a_model

SDXC

• Use 1024x1024
• First generation is really really slow, subsequent generations are fast

https://stable-diffusion-art.com/sdxl-model/#Download_and_install_SDXL_10_models

SD3.5

https://www.stablediffusiontutorials.com/2024/10/stable-diffusion-3-5.html

Flux AI

• You must use Stable Diffusion WebUI Forge
• In the UI, click "flux"
• First generation is really really really really slow

https://stable-diffusion-art.com/flux-forge/

This is working as of 5/12/25:

https://civitai.com/articles/9060/how-to-set-up-and-run-flux-on-forge-even-if-you-have-low-vram

Low VRAM

https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/blob/main/flux1-dev-bnb-nf4-v2.safetensors

Generate 512x512 images.

Nginx

htpasswd -c /etc/nginx/.htpasswd yourusername

sudo useradd -m -s /bin/bash nginx

sudo certbot --nginx -d reesimulate.reeseapps.com