Vision node

Person Segmentation

On-device person segmentation that turns any image into a silhouette alpha matte. It runs Apple's built-in segmentation models entirely on your Mac — nothing leaves the machine — and outputs a clean white, premultiplied mask of the subject that you can use for background removal, compositing, or as a mask feeding a Canvas.

Inputs

PortTypeDefault
ImageThe picture to cut a silhouette out of. Connect a Camera Input, or any other generator whose texture should be analysed.image—
EnableOff releases the vision request entirely, so nothing is analysed and no mask is produced.booltrue
ModeWhat to isolate — Person, Cat, Dog, or Subject for any prominent object. Clamped to those four.int0
QualityFast, Balanced or Accurate silhouette detail. Clamped to those three, and it only applies in Person mode.int1
Tracking FPSCaps how often frames are analysed. Auto (0) runs on every incoming frame; a lower cap saves a lot of GPU when the mask does not need to be crisp.int0
XHorizontal position of the mask in Grafit units, 0 being screen centre. Not clamped.float0
YVertical position of the mask in Grafit units, 0 being screen centre and positive going up. Not clamped.float0
ScaleUniform size multiplier for the mask. Not clamped: 0 collapses it, negative values mirror it on both axes.float1.0
RotationTurns the mask, in degrees clockwise. Any value works — it simply keeps rotating.float0
MirrorHorizontal flip, on by default so the silhouette matches a selfie-mirrored camera picture.booltrue
Flip YVertical flip, for sources whose frames arrive upside down.boolfalse
InvertSwaps subject and background, so the mask covers everything except the detected subject.boolfalse
OpacityFades the silhouette matte. Not clamped, but only 0…1 is meaningful.float1.0

Outputs

PortType
ImageThe picture to cut a silhouette out of. Connect a Camera Input, or any other generator whose texture should be analysed.image
Widthfloat
Heightfloat

Overview

Connect any image — a live Camera Input frame or any generator texture — and Person Segmentation isolates the subject from the background. The result is a white silhouette on transparency: where the subject is, the mask is opaque white; everywhere else it is transparent. Because it works on any image (not just a camera), you can segment pre-rendered footage, still images, or procedurally generated content just as easily as a webcam feed.

The Mode input chooses what to isolate. Person mode uses Apple's dedicated person segmenter and supports the Quality control for a speed/accuracy trade-off. Cat, Dog and the generic Subject modes use the foreground-instance segmenter to lift the most prominent matching object out of the scene.

Segmentation is aspect-correct on any source: the mask is produced at the source image's aspect ratio and aligned to where the image is drawn, so a downstream Canvas mask lines up exactly. A full transform block (X, Y, Scale, Rotation) lets you reposition the matte, and Mirror, Flip Y, Invert and Opacity give final control over orientation and strength.

Examples

  • Background removal — Segment a Camera Input in Person mode and use the matte as a Canvas mask to drop a new background behind the subject.
  • Pet cut-out — Switch Mode to Cat or Dog to lift a pet out of footage for a sticker-style overlay.
  • Generic subject isolation — Use Subject mode on any image to mask out the most prominent foreground object without naming it.
  • Glowing silhouette — Colour the white matte and feed it through a Bloom for a neon outline of the subject.
  • Invert for vignette — Turn on Invert to mask everything except the subject, e.g. to darken or blur the background.
  • GPU savings — Raise Tracking FPS on a near-static scene so segmentation runs less often while the matte stays in place.

Open Person Segmentation in the interactive catalog →

Person Segmentation is one of 381 nodes in Grafit Pro.

Free download. Every feature for 7 days, no card. Then €99.99 once — or keep playing your projects as a Player.

More Vision nodes