Installation¶

A modern NVIDIA GPU is required to run this application.

Anything less than 8GB of VRAM is not recommended, even that might be too little.

Multiple models are loaded into memory, so the more VRAM you have, the more models you can load.

On a 16gb GPU, you can load 3 models at once, maybe.

First clone the repository:

git clone -b release --single-branch git@github.com:BillSchumacher/HardDiffusion.git
cd HardDiffusion

Redis¶

Redis is required to run this application. You can install it on Ubuntu with:

sudo apt-get install redis-server

The easiest way to install redis-server on Windows is to use WSL2.

Warning

It’s important to not bind redis to all interfaces without a password.

Your system will likely be completely compromised if it is exposed to the internet.

Do NOT do expose an unsecured Redis server to the internet!

Note

If you want to use distributed processing, you will need to listen on an interface reachable by all machines.

As the warning above states, make sure you secure this with a password or username and password.

You can also use a firewall to restrict access to the redis server. This is highly recommended.

If you have spaces in your user profile path on Windows, you will probably want to fix that.

Python 3.9.13 is recommended. Python 3.10 is not supported yet.

To manage multiple python installations, pyenv is recommended:

On Windows you can use pyenv-win:

Note

pyenv-win is not without issues, but it does work.

virtualenvs seem to be broken at the moment.

ensure Path is configured correctly.

Installing python 3.9.13 with pyenv:

pyenv install 3.9.13
pyenv local 3.9.13

Installing requirements:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

By default SQLite is used, but you can use any database supported by Django.

If you want to use distributed processing you will likely want to use PostgreSQL.