Skip to content
BetaThis project is currently in its beta version.Updates might take a while to roll out.There's still plenty of work left to do.Thanks for your patience while we keep building.
BetaThis project is currently in its beta version.Updates might take a while to roll out.There's still plenty of work left to do.Thanks for your patience while we keep building.
mHumanTrackAImHumanTrackAI

Getting Started

This page walks you through the first few minutes with mHumanTrackAI: launching the app, reading the dashboard, calibrating with a T-pose, and choosing a mode.

First launch

When you start the app, a splash screen appears while the C# app launches the Python engine in the background. The app manages the engine for you — there is nothing to start manually.

On first run, if the RTMW model files are not already present, they are downloaded automatically. If your network blocks the download, see Model download blocked.

Panic at any time

Press Ctrl+Alt+P (a global hotkey) or the on-screen PANIC button to immediately stop all input injection — desktop and gaming — regardless of which app has focus. You can also hold both open palms for 2 seconds to toggle tracking.

The dashboard

The dashboard is your live view. It shows the camera preview with a real-time skeleton overlay (encoded as JPEG at ~12 fps, rate-decoupled from tracking so it never adds latency).

Three view modes:

  • Overlay — camera and skeleton superimposed.
  • Side by side — camera on the left, skeleton on the right.
  • Skeleton only — a privacy view that hides the camera feed entirely.

Status HUD reports live FPS, latency (ms), the current gesture and action names, the tracking source (rtm:cuda / rtm:cpu / mediapipe / dummy), camera FPS, loop FPS, per-stage timings (det / pose / lift ms), and a one-line bottleneck verdict.

Calibration

Open the Calibration screen and run the T-pose capture wizard:

  1. Stand in a T-pose (arms straight out, level with your shoulders).
  2. Click Capture.
  3. The engine measures your neutral joint baseline and saves it to the active profile.

The calibration view shows whether a T-pose has been captured. Calibration improves the accuracy of pose-relative gestures such as lean and crouch in Gaming Mode.

Choosing a mode

mHumanTrackAI has two modes, selectable from the main UI (and switchable per app — see Settings & Profiles):

  • Accessibility Mode — gestures drive the real cursor, clicks, scroll and keyboard. Best for hands-free desktop control. See the Accessibility gesture table.
  • Gaming Mode — body movement drives a virtual Xbox 360 controller. Requires the ViGEmBus driver. See the Gaming mapping.

Modes can switch automatically per application. Define a rule (process name → profile + mode) and mHumanTrackAI switches when that app becomes the foreground window (polled every second).