Flappy Bird, but the bird only flies when you smile.

Happy Bird is a feel-good twist on the classic Flappy Bird game. No tapping, no clicking — just smile to keep the bird in the air. Built with real-time face landmark detection and a machine learning pipeline, it's a simple yet powerful project that literally brings a smile to your face.

happy bird

Your webcam feed is processed live using MediaPipe Face Mesh, which extracts 468 facial landmarks per frame. A subset of lip landmarks is fed into a scikit-learn pipeline (scaler + Random Forest classifier) trained to distinguish smiling from neutral/non-smiling expressions. The prediction is streamed to a browser-based Flappy Bird clone via FastAPI — smile and the bird flaps; stop smiling and it falls.

Webcam → OpenCV → MediaPipe Face Mesh → Lip Landmarks
 → sklearn Pipeline (StandardScaler + RandomForestClassifier)
 → Smile / No Smile
 → FastAPI → Flappy Bird (HTML/JS)

Happy-Bird/
├── smile/ # Training images — smiling faces
├── non_smile/ # Training images — neutral faces
├── templates/ # HTML frontend (the game)
├── face_detection.py # MediaPipe landmark extraction
├── train.py # Dataset builder + model training
├── model.py # Pipeline definition
├── model.pkl # Trained and serialized model
├── lip_landmarks_dataset.csv # Extracted landmark features
├── fast.py # FastAPI app + webcam inference loop
├── smile.mp4 # Demo — smiling
└── nosmile.mp4 # Demo — not smiling

Prerequisites: Python 3.8+, a webcam, and a working smile.

# Clone the repo
git clone https://github.com/RijoSLal/Happy-Bird.git
cd Happy-Bird
# Install dependencies
pip install fastapi uvicorn opencv-python mediapipe scikit-learn numpy
# Run the app
uvicorn fast:app --reload

Then open http://localhost:8000 in your browser. Allow webcam access, face the camera, and smile to fly.

If you want to retrain the classifier on your own data:

# Step 1: Extract landmarks from the smile/ and non_smile/ image folders
python face_detection.py
# Step 2: Train the Random Forest pipeline and save model.pkl
python train.py

The pipeline applies standard scaling before classification, keeping things clean and portable via a single model.pkl file.

Most ML demos are impressive but impersonal. Happy Bird flips that — the model isn't just running in the background, it's you controlling something in real time. It turns a smile into an input device, which makes the whole experience oddly delightful.

Simple stack. Real-time inference. Zero buttons required.

MIT — fork it, smile at it, make it your own.