How to Run Stable Diffusion on macOS Without a GPU

1. Recommended Configuration for Running Stable Diffusion

Memory: at least 16 GB DDR4 or DDR5
Storage: at least 10 GB of free space
GPU: an NVIDIA card with at least 6 GB of VRAM

If your hardware falls short, you can also use one of the various optimized forks that support lower-spec hardware, but generation time will increase.

The current development machine is configured as:

2.9 GHz 8-Core Intel Core i7
16 GB 2666 MHz DDR4
250 GB SSD

Since there is no GPU, you need to wait a while longer when generating images.

2. Running Stable Diffusion on macOS

Install anaconda

1
brew install --cask anaconda

Configure PATH

1
echo 'export PATH=/usr/local/anaconda3/bin:$PATH' >> ~/.zshrc

Initialize conda for the shell

1
conda init zsh

Here you need to close the window and log into Terminal again.

Download and enter the project directory

1
2
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui

Create a Python environment

1
conda create -n stabel python=3.10.6

Activate the Python environment and install dependencies

1
2
conda activate stabel
pip3 install -r requirements_versions.txt

Download the model

Go to https://huggingface.co/CompVis/stable-diffusion-v-1-4-original and download the sd-v1-4.ckpt or sd-v1-4-full-ema.ckpt file, then place it under the models/Stable-diffusion directory. There are also many other models on huggingface you can download and use, and you can try them online too. For example https://huggingface.co/spaces/IDEA-CCNL/Taiyi-Stable-Diffusion-Chinese .

Modify the runtime parameters to skip the GPU check, see [1]

1
export COMMANDLINE_ARGS="--lowvram --precision full --no-half --skip-torch-cuda-test --listen --port 7860"

Run the project

1
python launch.py

Visit http://127.0.0.1:7860 locally to open the UI.

If you see an error like stderr: Running command git clone --filter=blob:none --quiet https://github.com/mlfoundations/open_clip.git at startup, you can edit launch.py, find the def prepare_environment() function, and change git+https://https://github.com/ to git+https://github.com/.

3. Text-to-Image Test

There are many writing techniques, sentence patterns, and modifiers for Prompts; Stable Diffusion also has many parameters you can adjust. But this post is mainly about running Stable Diffusion on macOS without a GPU, so here we just enter bird for the test. The generated image is shown below:

4. References

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/1742