Eric Guo's blog.cloud-mes.com

Hoping writing JS, Ruby & Rails and Go article, but fallback to DevOps note

Running Stable Diffusion v2.1 Base on the Mac Studio and macOS 13 Ventura

• Permalink

Stable Diffusion and MidJourney are the two of the most popular text-to-image models today.. MidJourney need to pay minimal $8 per month but Stable Diffusion is $9 per month but you can freely to run in local.

I have no NVidia GPU, but I have a Base model of Mac Studio, so I would like to try after Apple declared CoreML support Stable Diffusion.

Please make sure you have a good network and have upgraded macOS, aka macOS Ventura before beginning.

brew install miniconda

gh repo clone apple/coremltools

cd coremltools # just clone repo

conda init bash

./scripts/build.sh --python=3.10

xcode-select --install

Then running 

python -m python_coreml_stable_diffusion.pipeline --prompt "a photo of an astronaut riding a horse on mars" -i models/coreml-stable-diffusion-v1-4_original_packages -o output --compute-unit CPU_AND_GPU --seed 305

# If not running stable-diffusion-v1-4, the --model-version must be specified.

python -m python_coreml_stable_diffusion.pipeline --prompt "a photo of an astronaut riding a horse on mars" --compute-unit CPU_AND_GPU -o output --seed 1106 -i models/coreml-stable-diffusion-2-1-base_original_packages --model-version stabilityai/stable-diffusion-2-1-base

# Fix the model path if an error is reported.

subl /opt/homebrew/Caskroom/miniconda/base/lib/python3.10/site-packages/python_coreml_stable_diffusion/pipeline.py

subl /opt/homebrew/Caskroom/miniconda/base/lib/python3.10/site-packages/python_coreml_stable_diffusion/coreml_model.py

Comments