You filmed a street scene, a classroom, a protest, a school event. Before you post it, the faces of people who did not agree to be on camera need to go. Blurring them frame by frame by hand takes hours. The Blur Faces in Video tool finds and follows faces automatically, but automatic is not the same as perfect. This guide shows how to use it and, more importantly, how to check the result so that no face slips through.
How the tool finds faces
A face detector called YuNet runs in your browser through onnxruntime-web. It looks for faces about six times a second and follows each one between those looks, adding some margin so that quick movements stay covered. About 10 MB of detector files are downloaded the first time and cached. Your video is not uploaded: detection, blurring and encoding all happen on your device.
That design explains the weak spots. A detector can only find what it can recognise as a face. Anything that makes a face harder to recognise, such as a turned head, a very small face in the distance, a hand or scarf in front of it, or deep shadow, may be missed for a moment. The tracker usually bridges short gaps but not long ones.
Step by step
- Drop an MP4, MOV or WebM clip. The tool reads it and finds faces; wait for Finding and tracking faces to finish.
- You see a frame of the video with the faces hidden. Move the preview slider to look at other moments.
- Each tracked face is a small chip in the Faces row. Switch off a chip if that person should stay visible, for instance yourself or someone who agreed; use the remove button on a chip for a false detection, such as a poster or a pattern.
- Where a face is not covered, drag a box over it in the preview to add one.
- Choose the style and shape. Then press the button to render, wait for the progress bar, and download the MP4.
Which style to choose
| Style | Good for | Watch out |
|---|---|---|
| Blur | A natural look; strength 5–100% | A light blur can be reversed by some software or recognised by people who know the person; use a strong one. |
| Pixelate | A clear signal that a face is hidden; pixel size 5–100% | Small pixel sizes keep a recognisable outline. |
| Black | The most certain hiding | Looks heavy; fine for sensitive footage. |
| Emoji | Light-hearted videos | The emoji must be large enough to cover the whole face; raise the padding. |
The Padding around each face slider goes from 0 to 80%. More padding covers hair, ears and chin, and protects you when a face moves faster than the tracker. The shape can be an oval or a box; the box covers more.
The review checklist
Do this before you press export, not after. A post that shows one unblurred face has failed at the one thing it was for.
- Scrub the whole timeline slowly. Stop on every moment where people turn, walk in or leave the frame. These are the moments when faces are lost.
- Look at the edges of the frame. People who enter from the side show half a face first.
- Look at the background. Small faces in the distance are the most often missed. Add a box if one is recognisable.
- Check reflections. Windows, mirrors, glasses and screens in the shot can show a face that the detector does not treat as one.
- Check what else identifies a person: name badges, number plates, a distinctive tattoo, a card on a desk. This tool hides faces only.
- Watch the exported video all the way through, once, at normal speed, before it goes online.
Limits you need to know
- Length and size: up to 2 minutes and 500 MB on a computer; on a phone, only the first 30 seconds. Output is capped at 1080p on a computer and 720p on a phone. Longer footage has to be cut into pieces first.
- Browsers: recent Chrome, Edge and Safari on a computer. The tool needs WebCodecs video encoding, which Firefox and some older browsers do not offer yet.
- Time: every frame is redrawn, so processing takes about the length of the clip up to several times that, depending on the resolution and your computer.
- Frame rate: a phone video with variable frame rate is saved at a steady rate.
- Sound: the original audio is added back to the new MP4. If the video has none, or the format cannot be read, the result is silent and the page tells you. The voices in the video are not changed, so voices can still identify people.
After you export
If the file is too large for your platform, shrink it with the Video Compressor. The blur is part of the picture once the video is saved, so compressing does not bring faces back. For still pictures use the photo version of the tool, Blur Faces. To keep only the sound of a recording, use Video to Audio.
Blurring is not a legal clearance
Hiding faces reduces the risk of identifying people, but it does not decide whether you may film or publish a scene. Rules about filming people, minors and public events differ by country and platform, and a blurred face does not replace permission where permission is needed. Keep the original file safe, and publish only the blurred copy.
Frequently asked questions
Can the blur be removed later?
The blur is rendered into the picture, so the exported file does not contain the original faces. Your original file does, so protect it as well.
Why was a face not blurred?
Probably it was turned away, small, covered or in shadow. Scrub to that moment and drag a box over it.
Can I blur everyone except me?
Yes. Switch off the chip for your own face in the Faces row.
Is my video uploaded?
No. Only the detector files are downloaded, once.