ByteScope

Blur Faces

Find the faces in a batch of photos, check every picture yourself, then cover them with a solid block or a coarse mosaic — no blur on offer, because a light blur can be undone.

Your files never leave your browser — all processing is local.

No blur on offer — a blur can be undone

The page is called Blur Faces because that is what people search for, but it never blurs. A light blur over a face keeps most of the structure underneath, and that is enough to restore it or recognise it again. So there are two covers: a solid block that overwrites the pixels, and a mosaic coarse enough to pass the same safety check as the screenshot tool.

How to cover
Cover with
Shape

A block of flat colour. The pixels are overwritten, so there is nothing to undo.

Drop photos here, or press Ctrl+V

or click to browse — up to 50 images. Nothing is uploaded.

Every output is redrawn from pixels, so EXIF (capture time, camera, GPS position) is not carried over. Each file is checked before it is offered.

Remove location and camera data without covering anything? Remove EXIF

What this page cannot promise

  • Automatic detection misses faces. Small, turned-away, partly hidden or backlit faces can go undetected. Look at every photo before you download, and draw a box over any face it missed.
  • Faces on statues, posters and screens may get a box too. Tap the box to leave it alone.
  • Covering a face does not remove anything else that can identify someone: name tags, tattoos, number plates, landmarks in the background.
  • Photos are not uploaded. The 232 KB detection model is downloaded once; everything else happens in this tab.

Faces in a screenshot? Redact screenshot

About this tool

The page is called Blur Faces because that is what people search for — but it does not blur anything. A light blur over a face keeps most of the structure underneath, and that is enough for the face to be restored or recognised again; the smudge only defeats a glance. So the two covers on offer are a solid block, which overwrites the pixels outright, and a mosaic coarse enough to pass the same safety check the screenshot redaction tool applies. Detection runs in your browser with YuNet, a 232 KB model from OpenCV Zoo released under the MIT licence: drop in up to 50 photos, each one is scanned in turn, and every face it finds gets a box with its confidence score.

Check every photo before you download

Automatic detection misses faces. Small faces, faces turned away from the camera, faces partly hidden behind a hand or a hat, and faces in strong backlight can all go undetected — so the tool makes you look at every photo before it lets you download the whole batch. Step through the pictures one at a time: tap a box to leave that face uncovered, drag a box to move or resize it, and draw a new box over any face the detector skipped. Until every successfully processed photo has been on screen, the download-all button stays disabled and tells you how many are left to check.

Keep one face, cover the rest

The most common request is the inverse of redaction: a class photo or a sports-day picture where your own child should stay visible and everyone else should not. The keep-one-face mode does exactly that — tap the face to keep, and every other box on that photo switches to covered. Shape and colour are shared across the batch: rectangle or ellipse, a solid fill in black or a colour you pick, or the mosaic. The preview is drawn by the same code that writes the file, so the cover you see is the cover you get.

Nothing is uploaded, and EXIF goes too

The photos most worth hiding faces in are the ones you least want to send to a server first. This page is a static file with no API behind it: the model is fetched once, detection runs in a worker inside your tab, and the covered images are encoded by your browser — the network tab stays empty while you work. Each output is a fresh redraw, so EXIF is dropped along with it: capture time, camera model and GPS position do not travel with the file. PNG in gives PNG out; everything else is saved as JPEG, one at a time or as a ZIP. If the faces you need to hide are in screenshots, the screenshot redaction tool is the better fit; to strip metadata from photos you are not covering, use the EXIF remover.

Frequently asked questions

Will it find every face?

No, and the page is built around that. Faces that are small in the frame, turned to the side, partly covered, or lit from behind can all be missed, and the model has no way to tell you what it did not see. That is why the download-all button is disabled until you have looked at every picture: step through them, and draw a box by hand over anything the detector skipped. Treat the automatic pass as a first draft, not a verdict.

Why is there no blur option?

Because a light blur does not remove a face, it only makes it harder for a person to read at a glance; enough of the underlying structure survives for the face to be restored or matched. There is no blur strength the tool could honestly call safe, so it does not offer one. The solid block overwrites the pixels, and the mosaic is only allowed at block sizes coarse enough to pass the safety check — the same rule the screenshot redaction tool applies to text.

Does it remove EXIF and the GPS location?

Yes. Every output is a newly encoded image drawn from pixels, so none of the original tags come along: no capture time, no camera model, no GPS coordinates. This matters for exactly the photos this tool is for — a picture of children taken at a school carries the school's location in its metadata. Each saved file is checked for leftover tags before it is offered. If you want to strip metadata from photos without covering anything, the EXIF remover does just that.

Are my photos uploaded?

No. The photos are opened with the File API, the YuNet model runs in a web worker in your tab, and the covered images are encoded by your browser. The page is a static file with no server to receive uploads, so there is no upload request to find in the network tab — only the page itself and the 232 KB model being fetched, and from then on the work happens entirely inside the tab.

Does it work on a phone?

Yes, in any modern mobile browser, and the review step is built for touch: the layout works on a 375-pixel-wide screen, boxes are big enough to tap on or off, and you move or resize a box by dragging it or its corner handle. Detection runs on the phone itself, so a large batch takes longer than on a laptop — the per-photo progress and the cancel button keep you in control while it works.

How do I keep my own child's face and cover everyone else's?

Switch to keep-one-face mode and tap your child's face: that box turns to uncovered and every other box on the photo switches to covered. Then check the photo for faces the detector missed and draw boxes over them by hand — the mode only decides what happens to the boxes that exist. If your child appears in several photos, repeat it on each one; the choice is per picture, because the face to keep is a different box every time.

What does it box by mistake, and what does it never hide?

Faces on statues, posters, billboards and screens look like faces to the model, so they may get a box; tap it off and move on. Nothing else is touched: a name tag on a uniform, a tattoo, a licence plate or a recognisable landmark in the background all stay exactly as they were. If those matter for the photo you are posting, draw extra boxes over them by hand before you download.

Related tools