Scalecord

A discord bot which provides commands for upscaling images using PyTorch supported (.pth) models. Integrated with OpenModelDB to provide model metadata, automatically download models, and keep models and model data updated over time.

Requirements

• Poetry
• Python ^3.10
• An Nvidia GPU
• A discord bot token

Python Guide

1. Clone this repository somewhere on your system.
2. Install our requirements with poetry install.
3. In the config directory, duplicate the dist.env.yml file and rename it env.yml.
4. Create a bot by clicking the link above.
5. Invite it to your server via by going to "Oauth2" -> "URL Generator", generate a URL and navigate to it.
6. On the "Bot" page in your bot developer portal, click "Reset Token", copy that token. Paste it into the env.yml file for the DISCORD_BOT_TOKEN key.
7. Fill in any additional optional values in the env.yml, remove any you don't wish to use.
8. Launch the bot using poetry run scalecord

Docker Guide

Here's a boilerplate docker-compose.yml for the project:

version: "3.8"
services:
  scalecord:
    image: investigamer/scalecord:latest
    container_name: scalecord
    environment:
      - DISCORD_BOT_TOKEN=your_bot_token_here
      - DISCORD_GUILD_IDS=your_server_id_here,another_server_id,comma_separated
      - GITHUB_ACCESS_TOKEN=github_access_token_optional
      # You may want this line for GPUs with less memory
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    ports:
      - "5000:5000"
    volumes:
      # Map a directory to each: app/cache, app/config, app/models
      - "C:/windows/path/to/cache:/app/cache"
      - "linux/path/to/config:/app/config"
      - "macos/path/to/models:/app/models"
    restart: unless-stopped
    # Below is required for upscaling with Nvidia GPU
    deploy:
      resources:
        reservations:
          devices:
          - driver: nvidia
            count: 1
            capabilities: [gpu]

Where do I get models from?

OpenModelDB is highly recommended since this app synchronizes tags and information for any models sourced from OpenModelDB. There is a command which will attempt to download all models from OpenModelDB after the bot has gathered the latest OpenModelDB data.

# To gather data
scalecord update data

# To download models
scalecord update models

Please note that currently this will only download models that have a valid direct URL. We plan to support downloading from other sources like Google Drive and Mega.nz in the future. Models not on OpenModelDB will need to provide any related metadata manually (docs pending).

How do I define custom models?

Duplicate the dist.models.yml file, rename it models.yml. Here you can define custom models sourced from wherever you prefer.

Planned Features and Improvements

• Possibly generative AI (stable diffusion) integration
• Wider support for automated model downloads
• More environment control, commands, settings, etc
• More robust README and supporting documentation
• Support for AMD / CPU-only upscaling
• Support for NCNN / ONNX models
• ChaiNNer integration