Video Redaction (Automated Detection) – DekkoDEMS knowledge base

DekkoDEMS can scan video evidence for faces and black them out automatically, with all processing running on Dekko EdgeAI, entirely on your device - nothing is sent to a cloud service. The result is a separate redacted copy; the original recording is never altered.

This article covers the two AI detection tools: Find faces (scan the clip and choose which faces to black out) and Follow face with a black box (track one face through the clip). For drawing boxes yourself, trimming the clip, and muting audio, see Redacting Footage: Manual Black Boxes.

💡

Redact footage is an Early Preview feature, and face detection is an AI estimate. Faces can be missed, especially in dark, blurry, or fast-moving footage. Always step through the redacted preview and verify every face is covered before exporting or sharing.

Before you start

Redaction is available on any uploaded video file, transcribed or not. Open it from the file's ··· menu (or More dropdown) and choose Redact footage.

Face detection is very processor-intensive. Allow at least 10 seconds of processing per second of footage - a 5-minute clip can take an hour or more, depending on your device.

Test on a small portion first. Drag the timeline's Start and End handles to scope the clip down to a short section, run detection there, and check the results before committing to a longer run. Detection only scans the selected clip range.

AI detection sensitivity

Both tools use the AI detection sensitivity setting:

Standard - balanced speed and accuracy. Start here.

High - slower, but catches more. Switch to High if faces are missed.

Some trial and error may be needed depending on the content - lighting, motion, and camera angle all affect detection.

Find faces (scan the whole clip)

Use this when several people appear in the footage and you want to choose who gets blacked out:

Set the clip range, then choose Find faces. The scan runs on-device with a progress indicator ("Finding faces…") - you can cancel at any time.

[Screenshot: Find faces scan in progress]

When the scan finishes, every detected face is listed with a thumbnail and the time spans where it appears.

Tick/untick a face to toggle it. Checked = blacked out; you'll see the box on the video. Use Select: faces / all / none to toggle in bulk.

Missing someone? Use Re-scan at higher sensitivity.

Choose Put black boxes on selected to apply, or discard to throw the scan away.

Selected items become permanent black boxes in the exported copy.

Follow face with a black box (track one face)

Use this when one person needs covering as they move through the frame:

Move the playhead to a frame where the face is clearly visible, then choose Follow face with a black box.

Drag a box around the face on the video preview to start following.

Dekko EdgeAI tracks that face through the selected clip and adds a moving black box only where the face was found — the panel shows the followed face and its matched sections (for example "6 matched sections · 3:35–3:39").

Use Show individual boxes to inspect, hide, or remove specific boxes, or Remove all to start over.

Verify before you export

Detection is a starting point, not a finding. Before exporting:

Step through the preview frame by frame around each person — confirm the box stays on the face through movement, turns, and partial occlusion.

Check the filmstrip ("Final redacted output") for any frame where a face peeks out.

If anything is missed, re-scan at High sensitivity, extend the tracking, or add a manual box (see the Manual Black Boxes article).

The reviewer confirms; the model only suggests.

Export the redacted copy

Create redacted copy in case saves the redacted file alongside the original; Download redacted copy saves it to your device.

The output is a new, separate file (for example carchase.redacted.mp4) containing only the selected clip range, with the black boxes burned in permanently. The original recording is untouched.

You can also mute spoken details (names, addresses) in the Audio section before exporting — Mute from here or Remove all audio.

FAQ

Does the footage leave my device during face detection? No. Scanning, tracking, and rendering the redacted copy all run in your browser, on your hardware.

How long will detection take? Assume at least 10 seconds of processing per second of footage in the selected range. Scope the clip down to test on a small portion first.

A face wasn't detected — what now? Re-scan with sensitivity set to High. If it is still missed, cover it with Follow face with a black box or a manually drawn box.

Can I choose which detected faces are redacted? Yes. After a Find faces scan, tap faces to toggle them individually, or use Select: faces / all / none.

Are the black boxes reversible? Not in the exported copy — they are burned in permanently. The unredacted original stays in the case.

Does this change the original file? No. Redaction always produces a separate redacted copy.