No upload to any server = zero waiting

Blur Faces in Video, Automatically

Faces in your video are found and covered for you. Your file never leaves your device.

⚠️ Automatic detection is not perfect. Always watch the finished video yourself before you share it. Faces turned away from the camera, very small faces, and heavily blurred or dark shots can be missed. When processing finishes we list the moments where a face may have slipped through.

0 B uploaded — this page has not sent your files to any server?

Safe because there is no upload — a video that needs faces covered is usually a video you cannot hand to a stranger. This tool never sends it to a server. Finding the faces, covering them and saving the result all happen on your own machine.How this differs from upload-based sites — most converter sites upload your video or recording to a server, process it there, and promise to delete it after a while. Until that happens, your file is sitting on someone else's server. vidmoa has no upload step at all. The more sensitive the file — a recorded meeting, a phone call, family footage — the more that difference matters.

Blurring faces automatically: how do you catch the ones it missed?

This tool finds the faces in a clip, covers them with pixelation, an emoji or a blur, and saves a new MP4. Even the face-detection model runs inside your browser, so the video never leaves your device, and when it finishes it lists the moments where detection dropped out. Detection is not perfect, so checking that list and the result is part of the job.

A closer look

The work happens in two passes. The first pass decodes the whole video and only looks for faces: an open-source face-detection model checks a copy of the frame scaled to 640 px on the long edge, every third frame by default. Detected faces are linked into a track per person, and the frames in between are filled in from the positions before and after. When a face disappears, its last position stays covered for 10 more frames, with the box growing slightly. The second pass decodes the video again, covers those positions and encodes the result. Rotation data from iPhone portrait clips is applied first, so faces aren’t scanned lying on their side.

The detection threshold is deliberately low, so borderline cases get covered rather than skipped. Even so, the backs of heads, very small faces and heavily shaken or dark shots can be missed. That’s why stretches where a tracked face was lost, or where a face suddenly vanished, are listed under “A face may have been missed here”; clicking one jumps the result video to that moment. The face model (about 7 MB) and its runtime download once, and AAC audio is copied across without recompression.

How to use it

When it helps

Street footage before postingCover passers-by caught in the background of a vlog or on-location clip before it goes online.
Keeping the presenter visibleTick Keep some faces unblurred and a list of face thumbnails appears after detection. Mark the host as Leave visible, press Continue, and everyone else is covered.
School events and kids’ gamesShare a recital or sports-day video in a group chat or a public channel with other families’ children covered.
Submitting a clip as evidenceMake a copy of a report or complaint video in which uninvolved people can’t be identified, while your original file stays untouched.

Common snags

The same person appears twice in the list.Someone who leaves the frame and comes back is tracked as a new face. Use the time bar under each thumbnail, and if it’s the same person, tap both. A face caught only once for a moment is left out of the list as a likely false detection, but it is still covered.
A fast-moving face shows for a split second.By default faces are checked every third frame and the gaps are filled in, so sudden movement can slip through. Precise mode checks every frame and closes those gaps, at roughly three times the detection time. A wider Coverage setting also helps.
A warning appears when I lower the strength.A weak blur can let a face be recognized, so a Blur strength below 8 shows a warning, and Pixelate at 11 cells or more shows a similar one. If protecting identity is the goal, Pixelate at 10 cells or fewer, or an opaque emoji, is the safer choice.
It’s slow, or the sound is missing.Clips over 5 minutes or 300 MB show a warning, and trimming to the part you need first is much faster; keep the tab on screen, because hidden tabs are throttled. Audio that isn’t AAC can’t be copied, so the result is silent, and you’re told when you load the file. In Safari and on iPhone or iPad, detection runs without GPU acceleration for stability, which can take longer.

Tip Even if you don’t need to exclude anyone, trimming the clip down to the part you’ll publish first cuts both the processing time and the time you spend checking the result.

FAQ

Does it catch every face?

No. Front and side views are usually found, but backs of heads, very small faces, and heavily motion-blurred or dark shots can be missed. That is why we show you the moments where a face may have slipped through — click one to jump there. Please watch the whole result before you share it.

How long does it take?

It depends on the length of the video and on your device. When you pick a file we show an estimate based on its frame count — for a one-minute 30 fps clip that comes to roughly 50 seconds to 2 minutes 10 seconds. Slower devices can take longer, and precise mode takes longer still. Keep this tab visible while it runs — browsers throttle hidden tabs heavily.

Is my video uploaded?

No. Finding the faces and covering them both happen inside your browser, and your video is never uploaded. The page does need a connection to download the face-detection model and its processing scripts when you first use them. Don’t take our word for it — open DevTools → Network and watch.

Which files can I use?

MP4 and MOV, including recordings straight from an iPhone. The audio track is copied across untouched, so the sound quality is unchanged. For other formats, convert first — for example WebM to MP4.

Will the quality drop?

Covering faces means the video has to be rebuilt, so there is some re-compression. On the "High" setting the difference is hard to see. Videos longer than 1920 px on the long edge (1280 px on phones) are also scaled down to that size, because the maximum resolution a device can encode varies a lot from model to model.

Pixelate or blur — which is safer?

Pixelate. Even a strong blur can leave a trace of the original shape. If the video really must not identify anyone, choose Pixelate and move the "Mosaic size" slider down to fewer cells — fewer cells means bigger blocks and less detail.

When should I use precise mode?

By default a face is located every third frame and the frames in between are filled in from the ones around them. If people move fast, or faces keep disappearing and reappearing, turn on precise mode. It checks every frame and takes about three times as long.

Is it free?

Completely free. No limits, no watermark, no sign-in.