Vision node
Face Crop
Crops the source image — a live camera at full resolution, a video or any generator — around a chosen person's face and outputs it as an image, with full framing control — crop size, headroom, eye-level roll alignment, aspect, output size, edge handling, and hold-when-lost. Smoothed and gap-tolerant.
Inputs
| Port | Type | Default |
|---|---|---|
| ImageThe camera whose full-resolution frame is cropped. Cable the same Camera Input that feeds the tracking; nothing is drawn without it. | image | — |
| BodyVision Tracking's Body output, used only to match a face to a numbered person. Leave it empty when Person is Any. | json | — |
| FacesVision Tracking's Face output — the face boxes and eye points the crop follows. Required. | json | — |
| PersonWhich tracked person's face to follow. Any keeps following the face nearest the previous crop (largest face to start); a person number matches the face nearest that person's nose, and picks nothing if none is close enough. | int | -1 |
| ToleranceHow close a face must sit to the chosen person's nose to count as theirs, in screen fractions. At least 0.001; around 0.02…0.1 is reliable in a crowd. | float | 0.02 |
| Crop SizeHow much room around the detected face box to keep. At least 0.1, where 1 is a tight face and 1.5 a natural headshot; 3 and up becomes a half-body shot. | float | 1.5 |
| HeadroomShifts the frame up so the face sits lower, measured in crop heights. Not clamped; roughly -0.5…0.5 stays on the face. | float | 0 |
| Roll AlignRotates the crop so the eyes stay level while the head tilts. Off keeps the crop upright. | bool | false |
| SmoothingDamps the tracked crop rectangle with a critically damped spring plus a noise dead-band. Clamped to 0…1: 0 snaps to every detector update, 0.5 follows at ~2 Hz, 1 glides at ~0.35 Hz like a slow camera operator (never strobes, never overshoots). | float | 0.5 |
| AspectShape of the output frame: Square, Free (the face box's own proportions), or Fixed to a ratio you type below. | int | 0 |
| Aspect WWidth part of the fixed ratio, for example 16 in 16:9. Only used when Aspect is Fixed; values below 1 are treated as 1. | int | 16 |
| Aspect HHeight part of the fixed ratio, for example 9 in 16:9. Only used when Aspect is Fixed; values below 1 are treated as 1. | int | 9 |
| Output SizeFixed keeps the picture the same size whatever the distance; Track size lets the cutout grow as the person walks closer. Anything other than Track stays fixed. | int | 1 |
| EdgeWhat happens when the crop runs off the camera frame: Transparent leaves a hole, Clamp smears the border pixels. Anything other than Transparent clamps. | int | 0 |
| When LostWhat to show when the face disappears: Hold freezes on the last good crop, Transparent draws nothing. Anything other than Hold goes transparent. | int | 0 |
| MirrorFlips the cutout left to right, for the selfie-style view most webcam patches expect. | bool | true |
| OpacityFades the cutout on the canvas. Not clamped; 0…1 is the working range and higher values add nothing. | float | 1.0 |
| XHorizontal placement of the cutout on the canvas, in Grafit units where 0 is centre. Any value works, it simply travels off screen. | float | 0 |
| YVertical placement of the cutout, in Grafit units where 0 is centre and positive is up. Any value works. | float | 0 |
| ScaleSize of the cutout on the canvas. Not clamped; 1 fills the viewport height, 0 and below make it vanish. | float | 1.0 |
| RotationTurns the cutout on the canvas, in degrees clockwise. Any value works; it just keeps spinning. | float | 0 |
Outputs
| Port | Type |
|---|---|
| ImageThe camera whose full-resolution frame is cropped. Cable the same Camera Input that feeds the tracking; nothing is drawn without it. | image |
| Width | float |
| Height | float |
| Found | bool |
What It Does
Face Crop is an automatic camera operator for faces. It takes the full-resolution camera frame, finds a chosen person's face, and outputs a tightly-framed crop that follows them around — like a face-cam that pans and zooms by itself.
How It Works
Connect the camera to Image, and Vision Tracking's Body + Face so it knows where the face is. It then computes a crop rectangle around the face — sized by Crop Size, lifted by Headroom, and (optionally) rotated so the eyes are level (Roll Align). The result is smoothed every frame so it glides rather than jumps, and it "holds" briefly if tracking blinks out, so the shot never snaps to black.
Parameters
| Parameter | What it does |
|---|---|
| Crop Size | How tight the framing is around the face. |
| Headroom | Space above the head. |
| Roll Align | Rotate so the eyes are horizontal (level head shot). |
| Aspect | Square / Free / Fixed ratio (e.g. 16:9). |
| When Lost | Hold the last frame, or go transparent, when the face disappears. |
| Mirror / Opacity | Flip horizontally; overall transparency. |
Examples
- Auto face-cam — a hands-free webcam that always keeps the presenter centered.
- Reaction overlay — a small, stabilized head shot in the corner of a bigger scene.
- Multi-face wall — several Face Crop nodes (Person 0, 1, 2…) tiled on a Canvas.
Tips
- Turn on Roll Align for a professional, always-level head shot.
- Use Aspect = Fixed 16:9 to match a streaming layout.
- If the crop is jittery on a distant face, increase the camera Resolution so there are more real pixels on the face.
Open Face Crop in the interactive catalog →
Face Crop 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.