Body Tracking
Track a person's body pose, dense face mesh with blendshapes, hands, and gestures.
What it does
Body Tracking runs a single machine-learning model over an image to find a person: body pose, hands and face at once. The full set of pins is always present, body keypoints, left and right hand keypoints, face keypoints, a skeleton preview and per-point confidences, so downstream wiring survives a toggle being flipped. Regions that are switched off come through empty.
Dense Face gives a 478-point face mesh plus 52 expression blendshape coefficients, Dense Hands gives finer hand landmarks, and Gestures gives a per-hand gesture label. A One-Euro smoothing filter is on by default to settle per-frame jitter, and the skeleton preview draws over the image while the node is selected.
When to use it
A scene driven from a person on camera: puppeting a character, triggering effects from a pose, placing objects at a hand
Facial animation or expression data, from Dense Face and Blendshapes, as ARKit-style coefficients for a face rig
Hand gestures (open palm, fist, thumbs up, peace sign and similar) as a trigger, without a hand-built classifier
The face alone: Detector Source set to face-only, a lighter and more accurate face-only detector
| Parameter | Type | Default |
|---|---|---|
image | OneOf([FieldOf(Raster)]) | — |
enabled | Boolean | true |
detector_source | String | "whole_person" |
model_quality | String | "balanced" |
confidence_threshold | Number | 0.300 |
hand_confidence | Number | 0.200 |
face_confidence | Number | 0.200 |
min_detection_confidence | Number | 0.500 |
min_tracking_confidence | Number | 0.500 |
dense_face | Boolean | false |
blendshapes | Boolean | true |
dense_hands | Boolean | false |
gestures | Boolean | false |
smoothing | Boolean | true |
smoothing_min_cutoff | Number | 1 |
smoothing_beta | Number | 0.500 |
line_width | Number | 3 |
joint_radius | Number | 6 |
point_size | Number | 2 |
color | Color | rgba(0.00, 1.00, 0.00, 1.00) |
preview_regions | String | "all" |
opacity | Number | 1 |
Gotchas
Blendshapes only fire when Dense Face is also on. Blendshapes alone, with Dense Face off, outputs nothing.
Hand and face confidence thresholds default lower than the body threshold, because hands and faces are smaller and are occluded more easily. When gestures or face data drop out, lower Hand/Face Confidence rather than the main Confidence Threshold.
Lower Smoothing Min Cutoff for calmer tracking at rest, at the cost of more lag on quick movement. Raise Smoothing Beta to claw that lag back during fast motion.
If a Model Quality tier is not installed, the balanced model is used instead with no other sign of it. Check results when expecting the accurate tier.
Worked example
Add Body Tracking and wire a Camera or Image into its input.
Leave Detector Source on whole-person and Enabled on. A skeleton preview draws over the image.
Turn on Dense Face and Blendshapes to get facial expression coefficients for a face rig.
Turn on Dense Hands and Gestures to get finer hand tracking plus a canned gesture label per hand.
If tracking looks jittery at rest or laggy in motion, adjust Smoothing Min Cutoff and Smoothing Beta to taste.