Face Tracking
Face Tracking studies each video and works out, moment by moment, where faces are on screen, whether a person is present, and whether they are looking at the camera. This page covers running it, its options, and where the results go.
One thing to know up front: Face Tracking draws nothing on your preview. There are no boxes or overlays to watch. Its results are stored per file, and the reason to run it is that other tools use them. Camera Selection picks angles based on who is looking at the camera, and the cleanup tools can remove footage where nobody is.
Run Face Tracking
- Open the Process menu and click Face Tracking…. You can also right-click files in the media bin.
- Check the files you want in the dialog. Face-camera files are listed first and already checked. Files whose footage a previous run already covered start unchecked, with the note Already tracked.
- Click Run Face Tracking.
Face Tracking needs video. If the menu entry is greyed out, hover it: Requires video means the selection has none. No other tool needs to run first.
The run works in the background, and the status bar reads Tracking faces. Progress, cancelling, and failures work the same as every other tool. See Run an AI Tool.
Choose the options
Sample interval is a slider in milliseconds, from 50 to 250, set to 100 by default. Its hint explains it: "How often to sample face landmarks. Lower = denser sampling, slower."
Only analyze timeline content is a checkbox, off by default. When on, it "Skips portions of the source file no timeline clip uses. Padded by 2s around each used span." Switching the dialog to On Timeline turns this on for you, and switching back to Selected in Bin restores your previous choice.
Only analyze transcribed speech is a checkbox, off by default. It limits the run to the parts of a file where someone is talking, which is usually most of what you care about and a lot less footage to chew through. Its hint reads: "Analyzes only where a transcript sentence is present, in this file or any file synced to it, padded by 1s. Files with no transcript coverage are skipped. Requires Transcribe and Sync."
Because it works from the transcript, speech recorded on a separate audio device counts too, as long as that file is synced to the camera. Files with no transcript coverage are skipped rather than analyzed in full, and a note afterwards tells you how many, for example Face Tracking skipped 3 files with no transcript coverage. The run still counts as finished.
The checkbox is greyed out until you have run both Transcribe and Sync on the project, with the hint Run Transcribe and Sync first to enable this. Checking both tools in the same Batch Process run doesn't unlock it, since the transcripts don't exist yet when you set up the batch.
Turning both checkboxes on narrows the run to footage that is on the timeline and covered by speech.
Check the results
When the run finishes, the file's card in the media bin gains a small eye glyph with the tooltip Face-tracked. Click the card's meta row to open the Module Results popover. Its Face Tracking row summarizes the file, for example "2 identities · on screen 78% · looking at camera 41% of the video." A file you haven't processed reads "Not run yet."
If the run finds few or no faces, a warning appears titled "Face Tracking — Possibly not a face camera", with a line like "No face detected in NN% of frames for <file name>." The run still completed. The warning is a hint that you may have processed a screen recording or other faceless footage.
Put the results to work
Run Face Tracking before Camera Selection so camera choices can see who is present and looking. The cleanup tools use the same data to find and remove footage where nobody is looking at the camera.
Face Tracking finds faces, but it doesn't name them. That's the job of Find Characters. You can run the two in either order, and the app links their results either way.