How to blur faces in a video

Before and after: real footage

We ran two licensed stock clips through FaceBlur at its default settings: mosaic, 72% strength, and a 42% margin. The videos on the right are the files FaceBlur exported.

Test 1: one front-facing person

Source: Girl smiling, portrait in the library, used under the Mixkit Stock Video Free License.

Test 2: two coworkers turning toward the camera

Source: Two coworkers facing the camera by Theo Decker, used under the Pexels License.

The masks hold in these clips. That does not tell you how FaceBlur will handle a profile, crowd, reflection, dark scene, obstruction, or fast cut. See the demo notes and source licenses.

Upload, check, export

1. Choose a clip. FaceBlur accepts MP4, MOV, and WebM files up to 90 MB. The codec still has to work in your browser.
2. Wait for the masks. The face model starts after you select a file. Masks appear over the faces found in the current frame.
3. Play and scrub. Check the beginning, middle, end, and both sides of every cut. Count the visible faces and compare that number with the masks.
4. Choose a cover. Mosaic, soft blur, or solid cover applies to every face FaceBlur finds. Increase the margin if a mask sits too close to the face.
5. Check the audio. Turn on Remove audio if voices, names, or background speech identify someone.
6. Export. FaceBlur tries MP4 first, then WebM if your browser cannot encode MP4. It does not change the original file.
7. Watch the download. Review the file you will share, not only the preview.

What the preview is telling you

The counter shows how many masks FaceBlur is drawing at that moment. It is not the number of different people in the video, and it is not a confidence score. If three faces are visible and the counter says two, stop there and inspect the uncovered person. A mask may remain briefly when detection drops, so a matching count is useful but still not proof that every face is covered correctly.

Watch the shape as well as the count. The cover should stay centered on the face and extend past its edges during movement. Look for a mask that trails behind, jumps to another person, shrinks too tightly, or disappears during a turn. Pause on the first bad frame, then scrub a few seconds before and after it. That short range usually makes the failure easier to see than watching the whole clip at normal speed.

After analysis, you can adjust individual face tracks or add moving covers. A person may appear in more than one track.

Why a mask can move or disappear

The detector returns a face box for each frame. FaceBlur smooths that box against its recent position and holds the last mask for about three-quarters of a second when detection briefly drops. The hold reduces flicker, but it cannot predict a face that moves far from its last position or stays hidden. That is where a mask can lag, vanish, or attach to the wrong nearby face.

Margin only enlarges a box that already exists. Use it when the cover is centered but too tight. If there is no mask, or a mask jumps to the wrong person, raising the margin will only make the wrong box larger. Repeated tracking errors need a manual editor, not a stronger effect.

What each control changes

Strength changes what happens inside a mask. Margin changes its size. Neither control helps when the detector finds no face at all.

Frames worth a closer look

• Entrances and exits: check the frames before a mask appears and after it vanishes.
• Scene cuts: inspect both sides of the cut.
• Profiles and head turns: the detector has less face to work with.
• Obstructions: hands, phones, hair, helmets, masks, and other people can hide a face.
• Small and distant faces: zoom in and check background people.
• Fast motion: scrub around quick pans, running, and camera shake.
• Reflections and screens: inspect mirrors, windows, photos, televisions, and device displays.
• Other identifiers: look for tattoos, badges, documents, plates, clothing, locations, voices, and names.

With several people, follow each person through the clip instead of watching only the largest mask. See the multiple-face notes or use the redaction checklist.

If something goes wrong

See MP4 input and export for codec details, or mosaic settings for block size and margin.

Other identifying details

Covering faces does not remove voices, tattoos, names, uniforms, locations, or other identifying details. FaceBlur is not a legal or regulatory certification.

Common questions

Can it cover only one selected person?

After analysis, you can adjust individual face tracks or add moving covers. A person may appear in more than one track.

Does it remove audio automatically?

No. Audio is preserved unless Remove audio is selected before export.

Why can an MP4 input become a WebM output?

Your browser may be able to play the source without being able to encode AVC/MP4. FaceBlur uses WebM in that case.

Is the output watermarked?

No. The exported MP4 or WebM has no FaceBlur watermark.