Kim2091/upscale-bot

A bot for Discord that allows users to upscale images using a variety of super-resolution models through the spandrel library

Discord Upscaling Bot

A Discord bot that performs image upscaling using various super-resolution models. Users can easily upload images to this bot and have them upscaled with the model of their choice.

Features

• Supports multiple super-resolution models through spandrel
• Configurable settings for VRAM usage and processing limits
• Supports both .pth and .safetensors model formats
• Implements a customizable queuing system for handling multiple upscale requests
• Automatically adjusts tile size based on available VRAM
• Resize an image with typical scaling filters
• Fuzzy model name matching
• Alpha channel handling options: upscale, resize, or discard
• Detailed image information retrieval

Prerequisites

• Python 3.9 or higher
• CUDA-capable GPU
• Discord Bot Token
• Terminal program nvidia-smi installed on your system

Installation

1. Clone this repository:

   git clone https://github.com/Kim2091/upscale-bot.git
   cd upscale-bot
   

2. Install the latest PyTorch CUDA version for your system here: https://pytorch.org/get-started/locally/

3. Install the other required dependencies:

   pip install -r requirements.txt

4. Install ImageMagick

5. Open the config.ini file in the project root directory and replace YOUR_DISCORD_BOT_TOKEN and YOUR_DISCORD_USER_ID with your actual Discord bot token and user ID. Set this up at https://discord.com/developers/

   • The bot needs the bot scope (and maybe applications.commands), then:
     • Send Messages
     • Attach Files
     • View Channels
     • Use Slash Commands (maybe)

6. Place your models (.pth or .safetensors files) in the directory specified by ModelPath in the config file.

Usage

1. Start the bot:

   python upscale-bot.py
   

2. In a Discord channel where the bot is present, use the following commands:

   • To upscale an image:

     --upscale <model_name>
     

     Attach the image you want to upscale when sending this command.

   • To list available models:

     --models
     

   • To resize an image:

     --resize <scale_factor> <method>
     

   • To get detailed information about an image:

     --info
     

     Attach the image or provide an image URL when sending this command.

Configuration

You can adjust various settings in the config.ini file:

• Precision: Can be 'auto', 'fp16', 'bf16', or 'fp32'
• DefaultTileSize and MaxTileSize: Control the tile size for processing large images
• MaxTotalPixels: Maximum allowed input image size (width * height)
• VRAMSafetyMultiplier and AvailableVRAMUsageFraction: Fine-tune VRAM usage
• ThreadPoolWorkers: Number of worker threads for processing
• MaxConcurrentUpscales: Maximum number of concurrent upscale operations
• UpscaleTimeout and OtherStepTimeout: Timeouts for upscaling and other operations
• DefaultAlphaHandling: Default method for handling alpha channels

Security

• Never share your Discord bot token.
• The bot implements basic input validation.

Acknowledgements

• This bot uses the Spandrel library for loading and handling models.
• Thanks to @the-database for the benchmarks contained in vram_data_bffp16.md & vram_data_fp32.md
• Thanks to the Discord.py team for their excellent Discord API wrapper.
• The bot uses Wand (ImageMagick) for additional image processing capabilities.