A Windows companion for face tracking and optional controller features on a rooted Meta Quest Pro, based on Qpro-Enhanced-FT by n0tmast3r. It supports Virtual Desktop or Steam Link, USB or wireless ADB, tongue tracking with NVIDIA CUDA, experimental AMD ROCm support, and CPU fallback. Optional features include independent eye gaze, relative pupil animation, individual cheek puff and suck, eyebrow sensitivity, and experimental Virtual Desktop hands + controllers and Touch Pro thumb-rest input.
V3.0.0 is being prepared as a release candidate. See the V3.0.0 release notes for the changes and experimental features. The Releases page remains the place to download published builds.
Download QproCompanion.V3.0.0.zip from Releases when published. Extract it before opening QproFaceTracking.exe. GitHub's automatic source ZIP does not include the Hub executable or packaged models and tools.
Qpro Companion is the new public name. The executable, existing app/data folders and .qptonguemodel format retain their QproFaceTracking names for update compatibility. The legacy-named QproFaceTracking.V3.0.0.zip asset contains the same bytes for older updaters; use either ZIP, not both.
The latest VRCFaceTracking from Steam is required. If you used a Qpro version before V2.0, record and train a new tongue model. Working V2.0 or newer models can be exported and imported into V3.0.0 through Model manager.
Developer tongue model is the default and includes the accepted v15 visibility improvement from new clean-shaven recordings while preserving the original direction outputs. Mustachio is a separate, opt-in, highly experimental v9 extension trained with one bearded and moustached wearer. Broader wearer and facial-hair validation remain pending. Quick refinement and Focused training extend the model selected under Facial tracking > Lower-face tracking > Tongue model, including Mustachio; copies made from an experimental model retain that status. Full dataset trains a new personal model.
An active independent-gaze Magisk module supplies gaze itself. When detected, the Hub keeps Independent eye gaze + convergence off and disabled; skip Check gaze setup and Prepare gaze. Tongue, pupil dilation and eyebrow controls still work separately. After changing the Magisk module, reboot the headset and use Reconnect USB or Check headset compatibility to refresh its status.
The runnable release keeps the Hub and guide in its main folder. Development source, tests and build scripts are in src/. Building a package requires the .NET 10 SDK, portable Zig 0.15.2, an installed VRCFaceTracking, and binary assets from an extracted Qpro release. Source archives do not contain the runnable tools or model weights.
Build the native controller stage from the current source first; packaging verifies its source, resource and binary hashes. For a release containing both bundled models, supply separate approved Developer tongue model and Mustachio folders. Each must contain exactly one model's gate checkpoint, direction checkpoint and sanitized metadata; do not use a folder containing personal captures or several model pairs. AssetRoot alone does not select or copy these models.
Older release assets contain a Python installer instead of the required NuGet archive. The build downloads pinned official Python 3.12.10 when needed and verifies its SHA-256. For an offline build, supply python-runtime/python.3.12.10.nupkg in the source folder or asset root. Replace the example paths below with paths on your PC:
cd src
.\controller-input\build-native.ps1 -Zig 'C:\path\to\zig.exe'
$releaseBuild = @{
VrcftInstallDir = '<your VRCFaceTracking install folder>'
AssetRoot = '<extracted Qpro release>\QproRuntime'
ControllerInputAssetRoot = '.\artifacts\controller-native-build'
DeveloperModelRoot = '<approved Developer tongue model folder>'
ExperimentalTongueModelRoot = '<approved Mustachio model folder>'
}
.\build-release.ps1 @releaseBuildFor help, share relevant Hub Activity lines in the community Discord. The server is not owned by Fwooffy. Qpro Companion is unaffiliated with Meta, Virtual Desktop, VRCFaceTracking, and VRChat.
- Yeusepe / QFTPlus documented runtime interfaces and controller sensor layouts researched for the independently authored experimental hands/controller prototype.
- n0tmast3r created the original Qpro-Enhanced-FT project that this edition is based on.
- Fwooffy maintains this edition and wrote its beginner face tracking guide.
- Lumince and the Singularity contributors created the headset root project used by this workflow.
- danwillm and the LinkFT contributors documented the Steam Link tracking source used by Qpro's module.
- Fwooffy wrote the original beginner Quest root guide, and glorpette made and hosts its GitHub version.
The two bundled choices are Developer tongue model and Mustachio. Personal imports and refinements remain available. Camera-based cheek prediction and cheek calibration are not included. Native Individual cheek puff (1/0) and Individual cheek suck (1/0) controls separate the stronger side during a Qpro tracking session; turning them off, pressing Stop tracking or closing the Hub restores the streaming app's original cheek values.
Pupil image filtering can use a verified NVIDIA CUDA or AMD ROCm runtime. Geometry and smoothing still run on CPU, with automatic CPU fallback and backend/performance messages in Activity. The camera loop uses the latest frames and avoids encoding unused previews.