Will it work on photos with multiple faces?

Yes. The detector finds every face it can in one pass and blurs them all at once. If it misses any — profile shots and partially covered faces are the usual culprits — switch to Paint blur and brush over them yourself.

How much blur is actually enough?

As a rule of thumb, use a radius of about 10% of the face's width in pixels, and never go below 12. A face 150 pixels wide wants roughly 15; a large portrait face at 400 pixels wants 40. The default of 18 is fine for a face that fills a decent part of the frame, but it is not enough for a close-up.

Is blurring better than pixelating or a black box?

A black box is the most secure because it removes the pixels entirely. Heavy blur is next. Pixelation is the weakest of the three — each mosaic block preserves the average colour of the area beneath it, and that regular structure gives reconstruction methods something to work with. If you have used a light mosaic in the past, treat it as obscured rather than removed.

Can a blurred face be un-blurred?

Not from a heavily blurred image. The blur discards the fine detail rather than scrambling it, so there is nothing left to recover. AI tools that claim to "enhance" a blurred face are generating a plausible face, not restoring the real one. Light blur is a different story — it can leave enough structure behind to be matched against a known photo.

Can I undo a blur while editing?

Yes. Use Erase blur to lift blur from anywhere you painted by mistake, or hit Re-detect to start the automatic pass again. Once you download the file the blur is baked into the pixels permanently, and Filesmith keeps no copy of the original — so hold on to your source file.

Does this remove the GPS location from my photo?

No. Blurring changes pixels, not metadata. A photo can have every face blurred and still carry the exact coordinates where it was taken, plus the timestamp and camera serial number. Run the file through the EXIF Stripper afterwards if you are posting it publicly.

Do my photos get uploaded anywhere?

No. Detection and blurring both run on your own device. The only thing fetched from the network is the detection model itself, which loads once from a CDN. Your photo never leaves the browser tab, and closing the tab discards it.

Which file types can I use?

JPG, PNG and WEBP. The saved file keeps the format you started with, so a JPG in gives you a JPG out.

Blur Faces in Photos — Free, Private, No Upload

When you actually need to blur a face

A protest crowd. A street scene from a trip. Someone else's kids at a school concert. A screenshot of a group call you want to post about. A photo of an accident you're reporting. In each case the photo is worth sharing and one person in it never agreed to be published.

Blurring the face is the obvious fix. Finding a tool that does it without asking you to upload the photo first is the hard part — and if the photo is sensitive enough to need blurring, handing the unblurred original to an unknown server is a strange way to start. This tool does the whole job on your own machine.

How the detection works

The Filesmith face blur tool with a photo loaded, one face automatically detected and blurred, and the blur strength slider set to 37 — One face found on the automatic pass and blurred at strength 37. The brush tools sit next to the slider for anything the detector misses.

Face detection runs through face-api.js, using its tinyFaceDetector model. The model file loads once from a CDN into your browser tab, and from that point everything happens locally — your photo is read into a canvas, analysed on your own CPU, and blurred in place. No image data is sent anywhere.

The model is quick and dependable on faces that look roughly toward the camera. It gets less reliable as conditions get further from that, and it will regularly miss faces that are:

Turned more than about 45° to the side
Small in the frame — a crowd in the background, for instance
Partly hidden behind hats, sunglasses, hands or masks
Dimly lit, backlit, or low in contrast against the background

This is normal for a lightweight in-browser detector, and it's the reason the manual brush exists. Treat the automatic pass as a first draft that saves you time, not as a guarantee. Always check the result yourself before you save.

Using the tool

1

Drop your photo Any JPG, PNG or WEBP. Nothing is uploaded.
2

Let the automatic pass run The model loads, finds what it can, and blurs every face it detects straight away.
3

Set the blur strength The slider re-renders live, so you can judge the result as you drag it.
4

Paint over anything missed Switch to Paint blur. The brush has a soft edge, so it blends instead of leaving a hard rectangle.
5

Erase where you went too far Erase blur lifts it back off. Re-detect starts the automatic pass over.
6

Save Downloads in the format you started with. Your original file is never modified.

Tip

For a crowd, run the automatic pass first, then take a large brush at a low blur setting and sweep the background in one pass. The brush goes up to 120px, so a distant crowd takes a few seconds rather than a few minutes.

How much blur is enough?

This is the part most tools skip, and it's the part that decides whether the blur actually works. Blur strength has to scale with how large the face is in the image. A radius of 18 obliterates a face that's 120 pixels wide and barely softens one that's 500 pixels wide — same setting, completely different outcome.

The working rule: use a radius of roughly 10% of the face's width in pixels, and never go below 12.

Face width in the photo	Typical case	Blur setting
~80 px	Someone in the background of a wide shot	12–14
~150 px	A person mid-frame in a group photo	15–18
~300 px	Head-and-shoulders portrait	28–32
400 px and up	Tight close-up, or a large photo	40

One check that takes five seconds and is worth doing every time: open the saved file and view it at 100%. Phone screens and shrunken previews flatter a weak blur — anyone downloading the full-size image sees far more than you did. If you can still make out the shape of the eyes and mouth at full size, go higher and save again.

Blur, pixelate, or a black box?

All three hide a face on a casual glance. They are not equally strong.

A black box is the most secure, because it doesn't obscure the pixels — it replaces them. There's nothing underneath to work back to. It also looks blunt, which is why most people avoid it.

Heavy blur is close behind. A large-radius Gaussian blur throws away the fine detail rather than rearranging it. Nothing is left to reconstruct from. AI tools that claim to "enhance" or "unblur" a face are inventing a plausible face from a model, not recovering the real one — the information is genuinely gone.

Pixelation is the weakest of the three, and it's the one people reach for most. Each mosaic block stores the average colour of the region it covers, in a fixed, regular grid. That structure is predictable, which gives matching and reconstruction methods a foothold that a smooth blur doesn't offer. If you've used a light mosaic on something sensitive in the past, treat it as obscured rather than removed.

The other reason to prefer blur: it degrades gracefully. A blur that's slightly too weak still hides a lot. Pixelation at too large a block size looks obviously censored while sometimes revealing more than you'd assume.

A face isn't the only thing that identifies someone

Blurring faces and stopping there is the most common mistake. Before you post, look over the rest of the frame for:

Name badges, lanyards and ID cards — often legible once someone zooms in
License plates and vehicle details
House numbers, street signs and shopfronts that pin down a location
Distinctive tattoos, jewellery or hair, which identify a person as reliably as a face to anyone who knows them
Screens and monitors in the background, including anything reflected in glasses, windows or mirrors
Documents, whiteboards and sticky notes caught in the shot

The brush works on all of these — nothing about it is face-specific. Paint over anything you don't want published.

Then there's the part that isn't visible at all. Blurring changes pixels; it doesn't touch metadata. A photo can have every face perfectly blurred and still carry the GPS coordinates where it was taken, the timestamp, and your camera's serial number, all embedded in the file. If you're posting publicly, run the blurred file through the EXIF Stripper before it goes anywhere.

Worth knowing too: in the EU and UK, a photograph in which someone is identifiable counts as their personal data under GDPR, and several other jurisdictions take a similar line. Rules differ by country and by context, and none of this is legal advice — but "I blurred the faces" is a materially stronger position than not having done so.