Diffusion Bot

A Discord bot for interacting with diffusion models like Stable Diffusion XL 1.0:

Getting started

Local

Prerequisites:

• git
• Python
• Poetry
• Node

1. Clone the repo

   git clone [email protected]:hkiang01/diffusion-bot.git

2. Install dependencies

   API dependencies

   Configure api\pyproject.toml. If the GPU version of torch isn't compatible with your machine, change the dependency as follows:

   # pyproject.toml
   torch = { version = "^2.0.1" } # cpu
   # torch = { version = "^2.0.1", source = "cu118" } # gpu

   cd api
   poetry config virtualenvs.in-project true
   poetry install

   bot dependencies

   cd ../bot
   npm install

3. Configure the Discord bot. Follow the instructions for Creating your bot in the Discord developer portal and obtain:

   • the application's client id
   • the bot's token

   Then copy them in a file .env, created as a copy of .env.sample

   cp .env.sample .env

   # bot/.env
   API_URL=http://localhost:8000
   DISCORD_BOT_TOKEN=put_token_here
   DISCORD_OAUTH2_CLIENT_ID=put_client_id_here

4. Start the API

   cd ../api
   poetry run uvicorn api.main:app --port 8000

5. Start the bot

   cd ../bot
   npm run-script build
   npm run-script start

6. Follow the instructions for Adding your bot to servers

7. Have fun!

Dev Container

Prerequisites:

• Docker
• Visual Studio Code

1. Install the Dev Containers extension

2. Start the dev container by clicking "Open a Remote Window", then selecting "Reopen in Container"

   

   Note: If you don't have a GPU, comment out the following lines in .devcontainer/devcontainer.json:

   // "runArgs": [
   //     "--gpus",
   //     "all"
   // ],
   

Adding more models

The whole code base is extensible. If you implement another subclass of Model:

• the API automatically adds it as another model able to be used in the dropdown of /docs
• the tests in ./api/tests/models/model_test.py are automacially parameterized such that you can run predict() in the debugger without any additional code