This script allows you to install or run Automatic1111's Stable Diffusion WebUI with using venv and pyenv on your macOS. So, you no need to install conda.

• Installs and Runs Automatic1111's Stable Diffusion WebUI with using venv on your macOS
• MPS Support with Pytorch 1.13 stable

• MacBook Pro '13 2020, Apple M1 16GB RAM

• In the case of Apple M1 (MacBook Pro '13 2020) with using MPS, will produce 2.86s/it at fastest.
• In the case of Apple M1 (MacBook Pro '13 2020) with using CPU, will produce 5.36s/it at fastest.

• macOS 12.3 or later
• Apple Silicon Macs

• Xcode command line tools
• Homebrew
• pyenv via Homebrew

Other requirements (cmake protobuf rust python etc...) will be installed automatically while runnning the script.

# Install Xcode command line tools.
xcode-select --install

# Install Homebrew
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install pyenv
brew install pyenv

# Download this repo.
git clone https://github.com/SpinningSamurai/stable-diffusion-webui-mps_venv

# Change directory
cd stable-diffusion-webui-mps_venv

# Install requirements
./webui.sh -i

If you face some permissions error about webui.sh, try below.

# Give the file execute permissions
chmod +x ./webui.sh

# Try again
./webui.sh -i

After that, put your favorite model (.ckpt file) in right directly.

# Start WebUI
./webui.sh -s

# Install requirements
./webui.sh -i

# Clean up directory
./webui.sh -c

# Open source directory
./webui.sh -o

If you want to start WebUI with command line arguments, rewrite or add your favorites to line 241 in webui.sh

Note
I'm not original author of this script.

I downloaded the original one from github but its author and the repo seem to have disappeared. So I decided to upload my own modified one because he/she allowed redistribution in the script.

Here are the major changes from the original one.

• Changed Pytorch version from 1.13.0.dev20220922 (Nightly) to 1.13.0 (stable)
• To reduce RAM usage and increase system stability, I've added --medvram instead of opt-split-attention-v1 in command line arguments.
• Some optimizations and bugfixes.

In the end, it led iteration about 2x faster than original one.

Feel free to ask me when you have a issue or pull request. :9

• Some samplers don't work with using MPS. For example "Eular a".
• Stable Diffusion 2.0 models will not work.

• Automatic1111's Stable Diffusion WebUI
• A Original author of this script. Thank you.

This repository is under MIT license.