Face Track Preview
See exactly what the tracker thinks it found
- image
- face_tracks
- IMAGE
Face Track Preview is the smallest node in the Temporal Face Detailer pack and arguably the one that saves you the most render time. It takes the original frames plus a FACE_TRACKS object and redraws the tracking overlay as an image - the same drawing you'd get from the debug_overlay output of Face Detect + Track or the all-in-one detailer, but available anywhere in your graph.
Why it exists
The tracking nodes already have a debug_overlay output, so you might wonder what this adds. The answer: reuse. If you've saved a FACE_TRACKS object, routed it around, or wired it into a second-pass Tracked Face Detail, you no longer have the overlay handy - and re-running tracking just to see it again wastes time. Drop this node in wherever the tracks are, wire the frames and tracks in, and you get the overlay freshly drawn without touching the detector.
What you see per face: the detection box with confidence, the stabilized crop window with its anchor cross, facial landmarks, the track ID, gap frames that were interpolated through occlusion, and the rotation angle when align_rotation is on. The ID is the part you'll care about most in practice - it's the number you reference in a track_prompts override ("0: photo of johndoe person, detailed face"), and if IDs are jumping between faces, that's a tracking problem you want to catch before sampling, not after.
Inputs and output
Two inputs, one output, nothing to configure:
image- the frame batch (same frames the tracks were built from).face_tracks- theFACE_TRACKSobject.
One IMAGE out, ready for Preview Image or a Video Combine.
How to use it well
The pack's recommended loop: run the graph once with Face Detect + Track feeding this node (or just the debug_overlay output), check that every face has a stable id N box through the whole clip, and only then wire face_tracks into Tracked Face Detail and spend GPU time on sampling. If boxes flicker or drop out in the preview, fix the detector first - the README is emphatic that unstable detection is a root cause of flicker that no downstream blending can fix. This node is the cheapest way to check that.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| face_tracks | FACE_TRACKS | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |