TD DeepFace
TD_DeepFace is a TouchDesigner component library for integrating DeepFace face analysis into real-time visual workflows.
Run face recognition, age, gender, and emotion detection from a video stream or TOP input — useful for interactive installations, live AV, and experiments with machine vision inside TouchDesigner.
Features
- Face detection and recognition — detect and identify faces in a live camera feed or any TOP input.
- Facial attribute analysis — estimate age, gender, and dominant emotion (happy, sad, angry, surprised, neutral and more) for each detected face.
- Any TOP as a source — analyze a webcam, video file, NDI stream, or a rendered texture: anything that exists as a TOP in your network.
- Results back into the network — analysis output is available inside TouchDesigner, so face data can directly drive visuals, sound, or interaction logic.
- Built on DeepFace — uses the open-source DeepFace Python framework, which wraps state-of-the-art face recognition models (VGG-Face, FaceNet, ArcFace and others).
How it works
The component takes frames from a TOP input and passes them to DeepFace running in Python. DeepFace detects faces and returns structured data — identity match, bounding box, age, gender, and emotion scores — which the component exposes back to the TouchDesigner network for use in generative visuals, control signals, or data logging.
Requirements
- TouchDesigner with Python support
- The
deepfacePython package installed for TouchDesigner's Python environment - A video source (webcam, video file, or any TOP)
Installation steps and the ready-to-use component are in the GitHub repository.
Use cases
- Interactive installations — mirrors and screens that react to a visitor's face, age, or emotion.
- Live AV performance — driving generative visuals with the audience's or performer's facial expressions in real time.
- Machine vision experiments — prototyping computer-vision ideas inside TouchDesigner without leaving the node-based workflow.