Talon integration for Magic: The Gathering Arena (MTGA). Uses screen-based recognition to play cards and press buttons. Still requires some manual input for cards with targets and selecting blockers/attackers, but gets a good chunk of the way.

Caveats:

• Tuned on a 1440p screen, may need tweaking for other resolutions
• Has some edge cases with unusual card borders
• Can struggle if you have many cards in hand
• Still needs other interaction tools for blocks, individual attackers, and generally uncommon interactions
  • Pairs nicely with an eye-tracker for this

Short video of usage: https://youtu.be/hkBFbhc1mvg

• Clone repo to Talon user folder e.g. C:\Users\<username>\AppData\Roaming\talon
• Either install dependencies manually or install Anaconda 3
  • Manually
    • Open C:\Users\<username>\AppData\Roaming\talon\.venv\Scripts in command prompt
    • pip install "numpy>=1.21.4" "opencv-contrib-python==4.5.5.62" "opencv-python-headless==4.5.5.62" "pillow>=8.4.0" pyautogui
  • Anaconda 3
    • Create environment from provided yaml conda env create -f talon_env.yaml
      • This is allows side-loading tools without messing with the Talon virtual environment
• Reload Talon

• Consider turning off subtitles as unfortunately they appear over the cards and can't easily be moved
• Run MTGA in full-screen
  • Disable alternate card arts
  • Ensure there are no overlays in the way
• A special magic mode will be enable when selecting MTGA which disables other commands to improve detection
• Use highlight cards or keep highlighting cards to show overlays and ensure that you're playing the thing you want
  • Highlights will close automatically or in continuous mode stopped with stop highlighting cards
  • Use select card X to activate a card
  • Use play X to play a card
  • You can skip highlighting first by saying fast X
• Use done or ok to press any orange button
• See mtga.talon for full command list
• Note: Provides the mouse talon integration for mtga mode from the excellent knausj_talon

Fairly hacky approach to the optical recognition of the cards, but it runs decently fast and works most of the time.

• Aims to run fast (currently <0.1s locally)
  • Lets numpy/OpenCV do the majority of the heavy lifting
• Use MTGA's blue/yellow "playable" border to find the rough region
  • Left-most card has a darkened effect
  • Use region to strip out everything else
• Use black border to find individual cards
  • Will always be delineated by blue borders
  • Parametrise black border to catch most cases
  • Pull border coordinates to get a clickable position