Redact faces in video

Learning to redact faces in video is a core privacy skill. This how-to walks through a complete offline workflow: open a local file, detect faces, fix misses with manual boxes, track subjects, burn mosaic/blur/solid into an export, and review before anyone else sees it. For definitions of face redaction as a concept, see face redaction explained; for what a released “redacted face” means, see what is a redacted face. Start the tool path via face features or download Video Blackout.

Before you start

• Work from a copy; keep the original under your evidence or retention rules
• Use an approved Windows PC; prefer offline processing for sensitive clips
• Know who must remain identifiable (e.g., an officer or primary subject) vs who must be covered
• Plan time for audio ranges if names are spoken

Video Blackout does not upload your video for processing. After install, work stays on the machine—see offline redaction and the homepage.

Step 1 — Open the footage locally

Launch the desktop app and open the file from controlled storage. Confirm picture and sound. Note scene difficulty: crowds, hats, masks, side profiles, mirrors, and low light all increase miss risk.

Step 2 — Run face auto-detect

Use automatic face detection to propose regions. Treat results as a first pass. Crowded sidewalks and distant CCTV will leave gaps; that is expected. On-device assistance is described under AI video redaction.

Step 3 — Add and adjust manual boxes

Draw manual boxes on missed faces, partial faces, reflections, and faces behind glass. Enlarge boxes slightly so motion and compression do not peek identity at the edges. Remove false positives on posters or background photos if your policy requires those covered too—or leave them if they are not real persons in the incident.

Step 4 — Track across frames

Apply tracking so each region follows the person as they move. Scrub the timeline. When tracking drifts, correct the box and resume. For walking subjects in CCTV, tracking saves hours versus frame-by-frame drawing—see CCTV video redaction.

Step 5 — Choose mosaic, blur, or solid

Mosaic pixelates identity, blur softens features, solid blackout maximally conceals. Policy and aesthetics differ: some agencies prefer solid for clarity that content was withheld; others prefer mosaic for less visual shock. Ensure the effect is strong enough that identity cannot be inferred—especially on high-resolution exports.

Step 6 — Handle related identifiers

A face redaction can fail if a name tag, login screen, or nearby license plate remains. Add PII boxes and plate detection as needed (PII, plates). Mark audio mute/beep ranges manually for spoken names (audio)—Video Blackout does not auto-detect speech.

Step 7 — Export and QA

1. Export with effects burned in.
2. Play the export in a standard player (not only the redaction UI).
3. Spot-check beginning, middle, end, and any high-motion segments.
4. Verify that protected subjects stay covered and approved subjects stay visible if required.
5. Only then deliver or publish the release copy.

Tips for hard scenes

• Slow motion scrub for fights or crowds
• Watch mirrors, TV screens, and glossy doors
• Re-check after zoomed digital enhancement requests—enhancement can reveal missed faces
• Do not rely on “they are far away” at 4K

Practice on non-sensitive samples before production. For buying context, read choosing video redaction software. Ready to try? Download Video Blackout for Windows or visit videoredaction.io. More tutorials sit in guides.

Worked example: sidewalk bystanders

Suppose you have a two-minute phone clip of an altercation outside a store. The primary subjects must remain visible for the investigation; eight pedestrians must not. Open the file offline. Run face detect. You will likely catch most frontal walkers and miss a few profiles and a child partly behind an adult. Draw manual boxes on misses. Track each covered person through the frame. Choose a strong mosaic. Scrub at the moment a bus passes—motion blur can confuse tracking; correct immediately.

Next, mute the bystander who yells a victim's surname. Export. Play in a normal player on a large screen. If any covered face becomes readable when you pause, strengthen the effect and re-export. This is the loop professional teams live in. Definitions live in face redaction explained; outcome language in redacted face.

Keeping approved faces clear

Selective redaction requires discipline. Tag or remember which trackers belong to protected people. Do not accidentally cover the complainant you meant to show. When two faces overlap, you may need temporary larger boxes that you tighten after they separate. If policy says “officer faces remain visible,” verify helmet cams and reflective visors did not trigger unwanted covers.

Time-saving habits

• Fix tracking drift early; waiting until the end multiplies rework
• Group similar bystanders mentally so you notice when one exit is uncovered
• After major camera shakes, pause and reacquire boxes
• Keep a notepad of timecodes that felt risky for the QA pass
• Never skip the standard-player playback step

With habits in place, face redaction becomes predictable work instead of heroic improvisation. Keep the approved app handy via Download Video Blackout and the homepage.

Checklist you can print

1. Original preserved; working copy opened locally
2. Scope of who stays visible written down
3. Auto-detect run; misses boxed manually
4. Tracking verified through doors, crowds, and shakes
5. Effect strength checked on a large monitor
6. Related plates/PII/audio handled
7. Export burned in; standard player QA done
8. Peer review complete before release

Tape this checklist near the redaction PC. Habit beats memory after a long shift. For CCTV-shaped jobs, also skim CCTV video redaction.

Aftercare: storing and sharing the result

Place the redacted export in the disclosure folder, not beside the original under the same name. Use suffixes. Restrict ACL so interns cannot grab unredacted masters by mistake. When emailing, prefer secure file exchange over giant attachments. If a recipient asks for “just a little less blur,” reopen the project from the original with a revised scope—do not try to reverse a burned-in export.

Log the operator, date, software, and effect style. That log is invaluable when questions arise months later. Keep learning with other guides and the homepage.