stable-diffusion-docker

Example for running stable diffusion on the BYU supercomputer

For original Stable Diffusion repo and instructions see https://github.com/CompVis/stable-diffusion/.

Running Instructions

First clone the repo to the supercomputer login node:

git clone https://github.com/dcakagi/stable-diffusion-docker.git

Configure the supercomputer modules/settings:

module load charliecloud/0.26
export LC_ALL=en_US.utf8

Create the Docker image:

cd stable-diffusion-docker
ch-image build --force -t stablediff .

Build the Docker image to a tar file:

ch-builder2tar stablediff ${HOME}

Make the required directories for saving models for offline use (supercomputer runs disconnected from internet):

mkdir ${HOME}/stablediff_outputs
mkdir ${HOME}/stablediff_model
mkdir ${HOME}/tags

Untar and setup the container - this will run the stable diffusion once in order to download the necessary internal models: (Note: This may throw an error when running the stable diffusion about CUDA not working - this is fine, we just need the model to download)

bash setup_stablediff.sh

Run stable diffusion on the supercomputer compute node:

sbatch --output ./output_results.txt --mail-user YOUR_EMAIL_HERE --job-name "stablediff" run_tag.sh