This project is part of (and also associate to) the Project AIRI, we aim to build a LLM-driven VTuber like Neuro-sama (subscribe if you didn't!) if you are interested in, please do give it a try on live demo.
Who are we?
We are a group of currently non-funded talented people made up with computer scientists, experts in multi-modal fields, designers, product managers, and popular open source contributors who loves the goal of where we are heading now.
demo.mp4
| Name | Purpose | Base | 🤗 HuggingFace |
|---|---|---|---|
| Entities | Cards, Card Stack | YOLO11n | 🤗 Get Model |
| UI | Data panels, Scores, Buttons | YOLO11n | 🤗 Get Model |
| Unified | Entities + UI | YOLO11n | N/A |
| Basic | Multiple card types | Description | Crowded cards |
|---|---|---|---|
 |
 |
 |
 |
- Through script
pixi shell
python src/main.py- Through Notebook
Open src/notebooks/Train.ipynb and run all cells.
We trained this model on our own datasets labelled with n<1k images using Label Studio with YOLOv11n as the base model, it's available on HuggingFace as well: proj-airi/games-balatro-2024-entities-detection.
The training was performed on a single NVIDIA 4080Super GPU with 16GB VRAM, the loss optimized well and converged within set 2000 epochs.
git clone [email protected]:proj-airi/game-playing-ai-balatro.git
git lfs install
git submodule init
git submodule update
pixi installNote
For macOS, when running the UI automation and doing Game AI playing, you may need to give the terminal (or your IDE, in our case, Visual Studio Code) the permissions:
- Access to other apps
- Screen Recording
- Bypass private window picker
- Accessibility
After assigning the permissions, please restart your terminal or IDE to make it work.
If you find our works useful for your research, please consider citing:
@misc{proj_airi_game_ai_balatro_2025,
title = {Game AI - Balatro (2024, game)},
author = {Project AIRI Team, Neko Ayaka, Makito, Rainbow Bird},
howpublished = {\url{https://github.com/proj-airi/game-playing-ai-balatro}},
year = {2025}
}
