The VRAM requirements to run this model are very high. You will need a CUDA 11-capable NVIDIA GPU with 10GB+ VRAM to avoid out of memory errors. If you have such a system, then you can install the required packages on Pop!_OS with the following command. Note that if you have not previously set up docker, it will be required to add your user account to the docker group after installing these packages, and then restart.
sudo apt install cog-stable-diffusion nvidia-docker2
The model is hosted at https://huggingface.co/, and it will be necessary to create an account, get an access token, and get permission to download the model.
- Create an account at https://huggingface.co/
- Create an access token at https://huggingface.co/settings/tokens.
- Get permission to download and use the Stable Diffusion model
- Download the model with
stable-diffusion download {{ACCESS_TOKEN}}
Arguments to stable-diffusion are passed on to the cog command with a predefined cog configuration for stable diffusion. For available arguments and values, see the predict.py script. Invoking stable diffusion with a prompt input can be done like so:
stable-diffusion predict -i prompt="Text to execute in model"