Stable Diffusion: macOS install help

Stable Diffusion macOS install help is our guide for this kind of setup:

• Stable Diffusion image generator

• Apple macOS Monterey operating system, current version

• MacBook Pro with M1 ARM chip and 32GB free RAM, not x86-64 Intel chip, not less than 32GB free RAM

• Homebrew package manager

• Python programming language

• cmake build automation

• git source code manager

• protobuf data structures

• Anaconda data science platform

See our Stable Diffusion image gallery:

➤ https://github.com/joelparkerhenderson/stable-diffusion-image-gallery

Guide Goals

This guide is our notes about our preferred way of installing Stable Diffusion.

• If you want an easy way to try Stable Diffusion on macOS, then try this first: https://diffusionbee.com/

• If you want an easy way to try Stable Diffusion on macOS and other operating systems, then try this first: https://github.com/invoke-ai/InvokeAI

• If you want maximum configurability on macOS, and you willing to install a bunch of software pieces, then try this guide.

This guide is intentionally step-by-step, because we want to help more people be successful, including people who are trying these kinds of software installs for the first time.

• This guide is intentionally using steps that help multiple macOS users on the same system, because we want this guide to be able to help people who share a computer, such as teachers and students in schools.

• This guide is intentionally doing a "kitchen sink" approach of installing everything, because this can help our people with more capabilities as they try Stable Diffusion.

We welcome constructive feedback via GitHub issues, or pull requests, or email to [email protected].

Thanks

Thanks to many developers working on Stable Diffusion and help for it, and special thanks to:

➤ https://www.assemblyai.com/blog/how-to-run-stable-diffusion-locally-to-generate-images/

➤ https://github.com/lstein/stable-diffusion/blob/main/README-Mac-MPS.md

➤ https://news.ycombinator.com/item?id=32678664

Homebrew

Install:

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

Verify the path is typical:

% brew --prefix
/opt/homebrew

Verify your system can locate the program:

% which brew
/opt/homebrew/bin/brew

Verify the version is 3.5.10 or higher:

% brew --version
Homebrew 3.5.10

Do any brew upgrades:

% brew upgrade
% brew upgrade --cask --greedy

Verify brew is working:

% brew doctor

Python

Install:

brew install python

Verify the path is typical:

% brew --prefix python
/opt/homebrew/opt/[email protected]

Verify your system can locate the program:

% which python3
/opt/homebrew/bin/python3

Verify the version is 3.11 or higher:

% python3 --version
Python 3.11.3

cmake

Install:

% brew install cmake

Verify the path is typical:

% brew --prefix cmake 
/opt/homebrew/opt/cmake

Verify your system can locate the program:

% which cmake
/opt/homebrew/bin/cmake

Verify the version is 3.26 or higher:

% cmake --version
cmake version 3.26.3

git

Install:

% brew install git

Verify the path is typical:

% brew --prefix git  
/opt/homebrew/opt/git

Verify your system can locate the program:

% which git
/opt/homebrew/bin/git

Verify the version is 2.40 or higher:

% git --version
git version 2.40.0

protobuf

Install:

% brew install protobuf

Verify the path is typical:

% brew --prefix protobuf
/opt/homebrew/opt/protobuf

Anaconda

Install:

% brew install anaconda

Verify the version is 2023.03 or higher, and verify that the auto-update version is identical to the Caskroom version:

% brew info --cask anaconda
==> anaconda: 2023.03 (auto_updates)
…
/opt/homebrew/Caskroom/anaconda/2023.03 (304.8MB)
…

If the version is not 2023.03 or higher, or the auto-update version is not identical to the Caskroom version, then reinstall and reverify:

% brew reinstall --cask anaconda
…
==> Uninstalling Cask anaconda
…
==> Installing Cask anaconda
…
🍺  anaconda was successfully installed!

% brew info --cask anaconda

Create /opt directory

We like to install software such as anaconda within a system-wide top-level directory named "/opt", rather than within your personal user directory named "/Users/me" or similar.

This is because:

• The directory "/opt" is what brew does by default

• The "/opt" convention tends help with tutorials and troubleshooting as needed.

• The system-wide directory makes it easier for multiple system users to all use the same Anaconda installation; for us, this tends to helps teachers and students who need to use shared systems.

Ensure the system-wide top-level directory exists:

% mkdir -p /opt

In the next step, we will install anaconda to this directory:

/opt/anaconda3

Script

Anaconda uses an installation shell script that you must run.

Find the script:

% find /opt/homebrew/Caskroom/anaconda -name "*.sh" 

Output shows the shell script name, which may use different version numbers depending on when you install Anaconda:

/opt/homebrew/Caskroom/anaconda/2023.03/Anaconda3-2023.03-MacOSX-arm64.sh

Run the shell script via sudo and sh and the script name from above:

% sudo sh /opt/homebrew/Caskroom/anaconda/2023.03/Anaconda3-2023.03-MacOSX-arm64.sh

Your system should prompt you to type your password, then should run the shell script.

License

Anaconda prompts you to agree to the license:

Do you accept the license terms? [yes|no]

Type yes, then return.

Directory

Anaconda prompts you to choose an installation directory:

Anaconda3 will now be installed into this location:
/Users/…/anaconda3

  - Press ENTER to confirm the location
  - Press CTRL-C to abort the installation
  - Or specify a different location below

If you are generally a novice user, then use the Anaconda default directory, such as:

/Users/me/anaconda3

If you are generally an advanced user, then you may want to use our preferred naming convention, which installs many programs into the user's opt directory, such as:

/Users/me/opt/anaconda3

Type your installation directory, then return.

conda init

Anaconda prompts you to run this:

Do you wish the installer to initialize Anaconda3
by running conda init? [yes|no]

Type yes, then return.

Success

Anaconda installs and advises you:

For changes to take effect, close and re-open your current shell.

Close your current shell.

Open a new shell.

Verify the installation works by running Anaconda using its full path:

% /opt/anaconda3/condabin/conda --version
conda 23.1.0

Path

Export your preferred path such as:

export PATH="$PATH:/opt/anaconda3/condabin"

Optionally, you can append the Anaconda path to your environment, such as by editing whichever of these files you use, then closing your shell and opening a new shell:

• /etc/bashrc

• /etc/zshrc

• $HOME/.bashrc

• $HOME/.zshrc

Verify there is exactly one conda program on your path:

% which conda
/opt/anaconda3/condabin/conda

If you have more than one conda program on your path, then you'll need to be sure you're doing the rest of this guide using conda that you want.

• You can adjust your path, so it chooses the right conda program first.

• You can run conda via the full path /opt/anaconda3/condabin/conda.

• You can uninstall the other conda programs, if no one on your system needs them.

Verify the version is 23.1.0 or higher:

% conda --version
conda 23.1.0

Update

Update conda just in case it's changed recently:

% sudo conda update -n base -c defaults conda

Stable Diffusion weights file

Download the Stable Diffusion weights file; heads up that the file is somewhat large, more than 4GB.

Stable Diffusion verison 2

Download the Stable Diffision version 2 checkpoint file:

curl -O -L https://huggingface.co/stabilityai/stable-diffusion-2-base/resolve/main/512-base-ema.ckpt

We prefer to move the file into its own directory because that makes it easier to use among all the system users:

% sudo mkdir -p /opt/stable-diffusion-checkpoints/2.0/
% sudo mv 512-base-ema.ckpt /opt/stable-diffusion-checkpoints/2.0/

Stable Diffision version 1 - DEPRECATED

Download the Stable Diffision version 1 checkpoint weights file:

curl -L https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt > sd-v1-4.ckpt

We prefer to put the file in its own directory so it's easier to access for all our macOS users:

% sudo mkdir -p /opt/stable-diffusion-checkpoints/1.4/
% sudo mv sd-v1-4.ckpt /opt/stable-diffusion-checkpoints/1.4/

Stable Diffusion repository

This section comes from:

➤ https://github.com/lstein/stable-diffusion/blob/main/README-Mac-MPS.md

Install

Install:

% cd ~
% git clone https://github.com/lstein/stable-diffusion.git
% cd stable-diffusion

If you have previously installed the repository, then change into its directory, and reset the branch:

% git reset --hard origin/master.

Link

Link to the CKPT file:

% mkdir -p models/ldm/stable-diffusion-v1/
% PATH_TO_CKPT="/opt/stable-diffusion-ckpt"  # use your own directory that contains your CPKT file
% ln -sfn "$PATH_TO_CKPT/sd-v1-4.ckpt" models/ldm/stable-diffusion-v1/model.ckpt

Create conda environment

Create:

% CONDA_SUBDIR=osx-arm64 conda env create -f environment-mac.yaml

You should see output such as:

Collecting package metadata (repodata.json)

Restart your terminal.

Activate

Activate:

% conda activate ldm

Preload models

Preload python models:

% python3 scripts/preload_models.py

Heads up that the script does downloads that are somewhat large, more than 1GB.

You should set output such as:

preloading bert tokenizer...
preloading Kornia requirements ...
preloading CLIP model ...

Run dream

Run:

% python3 scripts/dream.py --full_precision

You should see output such as:

* Initializing, be patient...
…
* Initialization done! Awaiting your command (-h for help, 'q' to quit)
dream> 

Note: we use the option --full-precision, and not the option --half-precision. This is because the option --half-precision requires autocast and won't work.

Troubleshooting TypeAlias

If you get this error...

File "…/stable-diffusion/src/k-diffusion/k_diffusion/sampling.py", line 10, in <module>
    from typing import Optional, Callable, TypeAlias
ImportError: cannot import name 'TypeAlias' from 'typing' ("…/opt/anaconda3/envs/ldm/lib/python3.9/typing.py)

Then try editing this file:

src/k-diffusion/k_diffusion/sampling.py`

Change this line:

from typing import Optional, Callable, TypeAlias

To:

from typing import Optional, Callable
from typing_extensions import TypeAlias

➤ invoke-ai/InvokeAI#302

Generating

When you have the dream prompt:

dream>

Then you can enter your own command such as:

"photography of a cat on the moon" -s 20 -n 3 --sampler k_euler -W 384 -H 384

You should see output such as:

Generating…
Outputs:
outputs/img-samples/000001.2080708373.png …
outputs/img-samples/000001.1839445463.png …
outputs/img-samples/000001.2463981689.png …

You might see warning messages that you can ignore for now:

…/stable-diffusion/ldm/modules/embedding_manager.py:152: 
UserWarning: The operator 'aten::nonzero' is not currently supported 
on the MPS backend and will fall back to run on the CPU. 
This may have performance implications.

Your output files are now ready in this directory:

outputs/img-samples

Troubleshooting

Troubleshooting can start with running these command line diagnostics:

brew --version

python3 --version

cmake --version

git --version

conda --version

which conda

# This may take a while because it's searching your entire system
find / -regex '.*/bin/conda' -type f -perm +111 -print

# This may show unexpected results if you have more than one conda
CONDA_SUBDIR=osx-arm64 conda env create -f environment-mac.yaml

conda activate ldm

python3 scripts/dream.py --full_precision