Vision Pose
Detect body, hand, and face poses in images with Apple Vision.
What it does
Vision Pose finds people in an image and outputs their body joints, hand joints and face landmarks as a live skeleton, using Apple's on-device Vision framework (macOS only). Depending on the Detection mode it tracks a 17-joint body, 21 points per hand, 76 face landmarks, or all three at once.
The outputs are a ready-to-use skeleton of points and lines, plus raw landmark data, 2D screen positions, an estimated body height, a per-detection confidence, and a count of how many poses were found.
When to use it
A performer's body, hands or face tracked live from a camera or video
Joint positions driving a rig, particles or a motion effect
An on-screen skeleton overlay for a webcam feed
Hand landmarks for gesture-driven control, up to 4 hands
| Parameter | Type | Default |
|---|---|---|
image | OneOf([FieldOf(Raster)]) | — |
mode | String | "body" |
confidence_threshold | Number | 0.500 |
max_hands | Number | 2 |
enabled | Boolean | true |
line_width | Number | 3 |
joint_radius | Number | 6 |
color | Color | rgba(0.00, 1.00, 0.00, 1.00) |
opacity | Number | 1 |
Gotchas
This node uses Apple's Vision framework directly and runs on macOS only. It is not available on other platforms.
Jittery or dropping detections respond to Confidence Threshold: raise it to keep only the most reliable joints, or lower it to hold onto weaker and partial detections.
Max Hands only applies in the hand, body_hand and all modes. It does nothing in plain body or face detection.
Worked example
Wire a camera or video image into Vision Pose.
Set Mode to body_hand to track both a body skeleton and hand joints together.
Raise Max Hands if you expect more than two hands in frame.
Adjust Line Width, Joint Radius, and Color to style the skeleton overlay.
Read Confidence and Count downstream to gate effects on whether a pose was found.