How to run Stable Diffusion on Raspberry Pi 4
Example Output
a black tree with golden leaves painted by Monet, autumn
first of all i have to say that this is for sure not the most convinient way to run Stable Diffusion. The Raspberry's 8GB RAM are below the recommended 12GB minimum RAM. And the CPU is so slow that a 400x400 px image takes ~45 minutes to be ready.
so why we are doing this?
Becaus we can. And because both are great technologies. So consider this repository for educational purposes only and the joy of learning :)
Prerequisites
running on Debian GNU/Linux 11 (bullseye)
- Having cloned the Stable Diffusion model from Huggingface
i am using currently version 1.5 => https://huggingface.co/runwayml/stable-diffusion-v1-5
git lfs install
git clone https://huggingface.co/runwayml/stable-diffusion-v1-5
- Having Diffusers Library @0.7.x
https://pypi.org/project/diffusers/
pip install --upgrade diffusers
- Run Your Raspberry Pi Headless
As we need any bit of RAM we can get, i am running the Pi without GUI. You can enable this boot mode via raspi-config
Execute
with all that setup you should be able to run that code
python app.py
The most critical part i encountered while booting the model in the beginning. Mostly the process was killed here. Since the diffuser library supports the low_cpu_mem_usage parameter, i got it also running on my Pi
on peaks i measure ~6.3GB RAM usage
happy diffusing! š„³