the macos README is lacking a lot of steps

[slk333]

the macos README is basically a "go figure yourself", it really needs more work. It does not mention the following things, which are yet obviously required

• Installing Python
• Downloading the weights
• Installing Anaconda
• Installing protobuff rust

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

Also, there is no explanation on how to install GFPGAN and realESRGAN

[user799595]

Some more instructions here: https://news.ycombinator.com/item?id=32680385

[psychedelicious]

Here is a rough draft, I've tried to make it very clear and take you from a fresh macOS install to having SD running.

M1 Mac Installation

First get the weights checkpoint download started - it's big:

1. Sign up at https://huggingface.co
2. Accept the terms and click Access Repository: https://huggingface.co/CompVis/stable-diffusion-v-1-4-original
3. Download sd-v1-4.ckpt (4.27 GB) and note where you have saved it (probably the Downloads folder)

While that is downloading, open Terminal and run the following commands one at a time.

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

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

# install python 3, git, cmake, protobuf:
brew install python git cmake protobuf

# install miniconda (M1 arm64 version):
curl https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-arm64.sh -o Miniconda3-latest-MacOSX-arm64.sh
/bin/bash Miniconda3-latest-MacOSX-arm64.sh

# install rust:
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# clone the repo
git clone https://github.com/lstein/stable-diffusion.git
cd stable-diffusion

#
# wait until the checkpoint file has downloaded, then proceed
#

# create symlink to checkpoint
mkdir -p models/ldm/stable-diffusion-v1/
PATH_TO_CKPT="$HOME/Downloads"  # or wherever you saved sd-v1-4.ckpt
ln -s "$PATH_TO_CKPT/sd-v1-4.ckpt" models/ldm/stable-diffusion-v1/model.ckpt

# install packages
PIP_EXISTS_ACTION=w CONDA_SUBDIR=osx-arm64 conda env create -f environment-mac.yaml

# only need to do this once
python scripts/preload_models.py

# run SD!
python scripts/dream.py --full_precision  # half-precision requires autocast and won't work

[magnusviri]

@psychedelicious I like your instructions. Notes: you shouldn't need to brew install python git for this to work. Xcode includes both and these instructions don't use that version of python anyway. In fact, the instructions coudl be simplified by not doing the xcode-select because brew will install the command line tools also. Some people might prefer to install rust with brew. Is there a reason you are using their install script instead of brew? You also left out conda activate ldm. I'm working to incorporate your changes to the pull request I just made.

[psychedelicious]

@magnusviri Trust your judgement in including/amending any of what I wrote.

No reason for installing rust via script; it's just how I remembered installing it. Especially for less technical people, installing everything possible using brew sounds like a good idea - if something goes awry it's easier to fix stuff using a package manager!

[Beach77]

create symlink to checkpoint

mkdir -p models/ldm/stable-diffusion-v1/
PATH_TO_CKPT="$HOME/Downloads" # or wherever you saved sd-v1-4.ckpt
ln -s "$PATH_TO_CKPT/sd-v1-4.ckpt" models/ldm/stable-diffusion-v1/model.ckpt

Wow the original README wasted so much of my time, so evil. Thanks for making this. But there is still one place I don't understand which is the quoted part which I'll manage to figure out soon enough but could be helpful for others if it's made a little easier to understand.

EDIT: Managed to figure it out but don't remember how. No need for the thumbs down user799595 just because someone doesn't know how to figure out badly written instructions.

[user799595]

Slightly off-topic, but does anyone know the difference between these two files:
sd-v1-4.ckpt
sd-v1-4-full-ema.ckpt
I think the latter means "full Exponential Moving Average", which is a style of training apparently. This latter model is also about 2x the size.

Is one better than the other for inference?