DuckConvert Mascot
DuckConvert
Blog

Blur Faces in a Photo

Faces are found automatically and covered for good — blurred, pixelated or blacked out. Anything it misses, you draw over yourself. The photo never leaves your device.

It finds most faces. You handle the rest.

“Blur faces” sounds like a promise, so it is worth being plain about what the detector actually does. It finds faces that are roughly facing the camera and big enough to see. It misses profiles, people looking away, faces at the very edge of the frame, faces behind glass or a mask, and small faces in a crowd. On a photo of a demonstration it will find the people at the front and quietly ignore several at the back.

That is why nothing is applied until you have looked at it. Every region it found is outlined, every one can be removed with a click, and you can drag over anything it did not find — a face, a name badge, a house number, a screen. The covering does not care what is underneath.

Covered, not hidden

In most photo apps, a blur is a filter: the original pixels are still in the file and the effect sits on top of them. Here the covered pixels are replaced, and the file you download is built from the replaced ones. There is nothing underneath to recover because nothing underneath was carried forward.

The strength is a proportion of the face rather than a number of pixels, which matters more than it sounds. A fixed twelve-pixel blur erases someone standing at the back of a group photo and merely softens someone in a portrait, where every feature is still perfectly legible. Tying it to the region means the setting behaves the same at any distance from the camera.

The metadata goes too

A photo with the faces removed and the GPS coordinates still attached is not de-identified. The output here is encoded fresh from the pixels, so the camera, the timestamp and the location do not survive into it. If the picture itself is fine and only the metadata is the problem, the EXIF stripper removes it without re-encoding anything.

Nobody is told you opened this

The detector is a small model and a WebAssembly runtime, both served from this site rather than from the CDN the library reaches for by default. That default would mean a request to Google every time someone opens this page — a much smaller leak than the photo, but a leak, and an avoidable one. The photo itself never goes anywhere at all: there is no server in this tool to send it to.

Blur Faces FAQ

Will it find every face?

No, and you should not use it as though it will. It reliably finds faces looking roughly at the camera and reasonably large in frame. It misses profiles, faces turned away, faces partly out of frame, faces behind glass or a mask, very small faces in a crowd, and faces in low light. That is why everything found is shown to you before anything is applied and why you can drag over anything it missed. Treat it as a fast first pass that you check, not a guarantee.

Can the blur be reversed?

Not from the file you download. The covered pixels are replaced, and the output is built from the replaced pixels — there is no layer to peel off and no original underneath. This is the difference between this and a photo app where 'blur' is a filter you can toggle: there the original is still in the file. At the default strength the blur radius is a third of the face's width, which is far past the point where anything can be recovered from it. If you want to leave no doubt at all, use the solid block.

Is my photo uploaded to be scanned for faces?

No. The face detector is a sub-megabyte model your browser downloads once from this site and then runs on your own device. The photo is read by the page and never sent anywhere. That matters more here than almost anywhere else: the photos worth de-identifying — protesters, children, patients, bystanders — are exactly the ones you should not be handing to a stranger's server in order to have the faces removed.

Does the downloaded photo still have GPS data in it?

No. The output is encoded fresh from the pixels, so the camera model, the timestamp and above all the GPS position do not travel with it. That is not a bonus: a photo with the faces removed and the coordinates of the front door still attached is not de-identified, and a tool that produced one would be actively misleading. If you want to strip metadata without touching the picture, the EXIF stripper does exactly that and nothing else.

Which is stronger — blur, pixelate or a solid block?

A solid block, obviously, and it is the only one where there is nothing to argue about. Between the other two, coarse pixelation and a heavy blur are both effectively irreversible at the sizes used here, because the strength scales with the face rather than being a fixed number of pixels — 'strong' means the same thing on a face 40 pixels across and one 400 across. Pixelation looks deliberate, which is sometimes what you want; a blur looks less like a redaction.

Can I use it on something other than a face?

Yes. Drag over anything — a name badge, a house number, a licence plate, a screen. Those regions are covered exactly the same way and just as permanently. The detector only looks for faces, but the covering does not care what is underneath it.