FaceFlow – Face Tracking, Recognition & Face Swap

Face tracking, participant recognition, expressions and face swapping for Unity on Windows. Includes an editable Studio scene, webcam and video input, C# source and setup guides.

PLATFORM

Windows x64. Tested player scripting backend: Mono. Unity 2022.3.21f1 and Unity 6000.0.67f1. Other platforms and IL2CPP are not validated.

INFERENCE

ONNX Runtime with DirectML GPU inference or CPU processing. Optional CPU fallback handles model startup and runtime frame-processing failures.

Bundled family: YuNet detection, SFace recognition and MobileFaceNet expression classification.

InsightFace family: SCRFD detection, ArcFace recognition and InSwapper face swapping; external model files and separate model terms apply.

INPUT AND OUTPUT

Webcam input and local H.264 MP4 video through Unity VideoPlayer. Video processing preserves aspect ratio up to a 1920 × 1080 frame. Video loops without audio.

Portrait input: PNG, JPEG and still WebP. Processed output can be saved as PNG. Processed-video recording is not included.

INTEGRATION

FaceFlowCamera component, per-face observations, recognition and presence events, and ScriptableObject settings. Editable Studio scene and prefab use Unity UI and TextMeshPro. Input System is optional.

Participant enrollment is held in memory; it is not saved across application restarts. Track IDs are temporary. Expression labels are model estimates, not a measurement of a person’s internal emotional state.

DOCUMENTATION AND SUPPORT

Included: Documentation/FaceFlow Guide.html, Getting Started.md, API Reference.md and Integration Examples.md.

Support: hello@reother.com

Website: https://reother.com

Build a face-driven experience without wiring the whole pipeline yourself.

FaceFlow brings face detection, tracking, participant recognition, expression classification and face swapping into one Unity toolkit. Start with the included Studio scene, then adapt its UI or use the runtime components in your own project.

WHAT YOU GET

• Webcam and imported video input.

• Multiple-face tracking with automatic largest-face selection or manual selection.

• Guided participant enrollment with multiple reference angles. Completed participants stay available when switching sources during the session.

• Seven expression categories with smoothing and an Uncertain state when the result is unclear.

• Face swapping from a portrait, adjustable blending and PNG output.

• An editable TextMeshPro interface, C# source, integration examples and an offline guide.

MODELS AND FACE SWAPPING

Detection, recognition and expression models are included. Face swapping requires separately downloaded InsightFace models. Setup provides the download and model checks. The public InsightFace weights are for non-commercial research; commercial use requires an appropriate licence from InsightFace. Purchasing FaceFlow does not grant that model licence.

Built for Windows desktop projects, interactive installations and prototypes. Processing runs locally; no Python installation or inference service is required. Speed and accuracy depend on hardware, lighting, face size and the input footage. Use footage and portraits you have permission to process.