Blur Faces in Video

Find the faces in a video automatically and blur, pixelate, cover or permanently destroy them, with detection running on your device and no upload.

Choose a video

or drop it here

The video is processed on this device and never uploaded.

MP4, MOV, WebM, MKV, AVI, WMV, FLV, 3GP, TS / MTS / M2TS or VOB / MPG

Files over 1 GB can be slow. Very large outputs need a browser that can save straight to a file.

How it works

Choose a video and press Find faces. YuNet, a small open-source face detection model, runs in your browser and checks the frames one by one. Nothing is uploaded. On first use the browser downloads the detection runtime and the model, about 7 MB compressed, and keeps them for offline use.

Each face is followed from frame to frame and shown on the preview in its own colour. Before you apply, delete anything that is not a face, change the mode or time range of a face if needed, and add a region for any face that was missed. When you press Apply, your browser draws the effect into every frame and encodes a new file on your device. The page then re-reads that file and checks every face and region.

Blur looks the most natural, but a blurred face can sometimes be partly recovered, or recognised by someone who knows the person.
Destroy scrambles the face into a few large shuffled blocks that change every frame, flattens them and then blurs the result, so the original cannot be reconstructed. Use it when the person must never be identifiable.
Pixelate gives a mosaic and Cover a solid ellipse.

Supported files

Inputs: MP4, MOV, M4V and WebM in Chrome, Edge and Safari on a computer (face detection needs WebCodecs). MKV works in most desktop browsers. AVI and FLV files with H.264 video are repacked with FFmpeg first (copied, not re-encoded); WMV and VOB are not supported here yet.
Outputs: MP4 (H.264 and AAC) or WebM (VP9 and Opus), at high quality and the original resolution.

Limitations

Detection can miss faces that are smaller than about 1/40 of the frame, in profile or turned away, partly hidden, in the dark or blurred by motion. Review the faces before you apply, and add a region for any that were missed.
It can also mark things that are not faces, such as posters, statues or patterns. Remove those with Not a face in the list.
One video can take at most 64 faces and regions. In a larger crowd, small or short faces are merged with a neighbour into one larger box, and any that cannot be merged are listed and left visible.
Blur and pixelation can sometimes be partly reversed, and people can still be recognised by their hair, clothes or voice. Use Destroy when the person must never be identifiable.
The video is always re-encoded, so it takes longer than a copy and loses a little quality. HDR videos come out as standard dynamic range.
Browsers without WebCodecs use the FFmpeg fallback. It draws ellipses as rectangles with hard edges, and Destroy becomes a solid black box, which is still unrecoverable.
The sound is not changed: voices and names said aloud stay in the file. Use Remove Audio if the sound must go.
Face detection needs WebCodecs, so it does not work in Firefox on Android or in older browsers. Blur Video, where you draw the boxes yourself, works in more browsers.

Questions

Is my video uploaded for face detection?

No. The detection model runs in your browser on your device. The only download is the detection runtime and the model (about 7 MB) on first use, and the browser then caches them.

Will it find every face?

Not always. Small, distant, turned or hidden faces can be missed. Thorough detection finds smaller faces but takes 2 to 4 times longer. Review the faces on the preview before you apply, and add a region for any that were missed.

Should I use Blur or Destroy?

Blur looks softer, but a blurred face can sometimes be partly recovered or still recognised. Destroy shuffles and flattens the face before blurring it, so there is nothing left to recover. Choose Destroy when the person must never be identifiable.

What happens with a crowd?

A video can carry at most 64 faces and regions. Beyond that, the smallest and shortest faces are merged with a neighbour into one box, and the page lists any face that could not be merged.

Is the sound changed?

No. Only the picture is changed, so voices and names said aloud stay in the file. Use Remove Audio to drop the sound.

Example

A 20-second street interview has two passers-by in the background. Press Find faces, check that the three faces are outlined, press Not a face on a poster that was picked up as a face, choose Destroy and press Apply. The output is the same clip with the interviewee and both passers-by scrambled. The sound stays as it was.