Open gesture-recognition framework

Teach a machine your gestures — from your webcam.

Capture hand, pose, and full-body gestures with a desktop app, then train and run recognition models with one command. No cloud, no labelling service, your data stays on your disk.

Download Gesto Labeller Read the quickstart

Gesto Labeller 1.0 · Community Edition gesto 0.2 · Python package

The framework

Two pieces, one workflow

Gesto splits the job in two: a friendly desktop app for making datasets, and a Python package that turns those datasets into working models.

Gesto Labeller Community Edition

Point your webcam, hold or perform a gesture, and capture labelled samples. Think "Teachable Machine for gestures," running entirely offline — free and open to everyone.

Capture static poses or motion sequences
Hands, two hands, pose, legs, or full body
Explore, relabel, and trim samples visually
Exports plain .npy files — yours to keep

gesto python package

Train a model from a Labeller project and run live detection — from the command line or a few lines of Python.

gesto train — static or sequence models
gesto detect — webcam or video, live bars
gesto image — classify a single photo
Versioned model storage, nothing overwritten

How it works

From webcam to working model

The same landmarks flow through every stage — captured, normalized, and matched identically at training and detection, so what you record is what the model learns.

Capture

Record labelled gesture samples from your webcam in Gesto Labeller.

Landmarks

MediaPipe extracts hand and body keypoints, normalized for position and scale.

Train

gesto train fits a static or sequence model and versions the result.

Detect

gesto detect runs it live on your camera or a video file.