Blur anything in a photo, without uploading it

What this is

BlurMyPhotos covers the parts of a photo you would rather not send out with the rest of it: a face, a license plate, a house number, a name on a letter, a balance in a screenshot. You open the image, black bars go over what matters, you download the result. There is nothing to install and nothing to sign up for, and on a decent photo the whole job takes under a minute.

Faces are found for you. Everything else you cover yourself by dragging a box over it. That split is the honest description of what the software can do, and the sections below explain where the line falls and why it falls there rather than somewhere more flattering.

Why nothing is uploaded

Most image tools online work by sending your file to a server, doing the work there and sending a result back. Many of them promise to delete it afterwards, usually within the hour. That promise can be entirely sincere and still not help you, because it is a promise about a copy of your photo sitting on a disk you do not control.

This site does not offer a better promise. It removes the thing the promise was about. Your file is read by the browser, decoded into pixels inside the page, edited in a canvas element and encoded back out to a file your browser saves. None of that touches a network. There is no upload endpoint, because the site is static files with no server-side code behind them.

A server that never receives your photo cannot leak it in a breach, cannot lose it in a misconfigured bucket, cannot be compelled to hand it over, and cannot keep a copy after you believed it was deleted. There is nothing to delete because there was never anything stored. If you would rather check that than take our word for it, how it works sets out four tests you can run yourself, including watching the network tab while you use the tool and finishing a photo in aeroplane mode (airplane mode).

The four steps

1. Choose a photo. Drop it on the box above, or tap it to open your camera roll. Nothing is sent when you do this; the browser simply hands the page a file it already has.
2. Faces are found automatically. Once you have picked a file, and never before, Google's MediaPipe BlazeFace model is downloaded and run on your own processor. It returns a rectangle for each face it recognises.
3. Adjust, and add what it missed. Move and resize the boxes, delete any that are wrong, and drag fresh boxes over plates, signage, text or anything else. Detection is faces only, so this step is where the rest gets covered.
4. Download. PNG at the original resolution, or JPG. No watermark, no account, no cap on how many photos you run through it.

What you can cover

Seven starting points. Six share the same image editor underneath; the seventh works on documents rather than pictures. Pick whichever matches the thing in front of you.

• Blur faces in a photo

  The only one here with automatic detection. Group shots, a school play, strangers in the background of a street scene.

  Automatic · bar, pixelate or blur

• Blur a license plate

  Before a car advert goes online. Drag one box over the plate and the characters are replaced rather than smudged.

  One box · irreversible

• Pixelate part of a photo

  Starts in pixelate mode, for when the shape of a thing should stay readable but the detail of it should not.

  Pixelate · adjustable block size

• Black out text in an image

  An address on an envelope, a name on a form, a booking reference. Solid bars, with nothing left underneath them.

  Solid bar · nothing recoverable

• Blur a screenshot

  An email address, a chat name, a bank balance — covered before the grab goes into a thread or a support ticket.

  Screens · PNG in, PNG out

• Redact a PDF

  A document rather than a picture. Every page is flattened, which is the only way a black box over PDF text is worth anything.

  Up to 120 pages · no text layer left

• Strip photo metadata

  No bars at all. It reads what your file is carrying, GPS coordinates included, and writes you a clean copy of it.

  EXIF · GPS · camera serial

There is no automatic licence-plate detection and no automatic text detection on this site. That is not an oversight. No reliable, permissively licensed model for either runs in a browser today, and shipping a detector that quietly misses half the plates would be worse than asking you to drag one rectangle. Drawing a box takes about two seconds and you can see that it landed.

Which mode to use

Solid bar

Paints the rectangle flat black, replacing the pixels. Nothing of the original survives in the exported file. Irreversible, and the default here.

Pixelate

Averages each block of pixels down to a single value. Those averages are real information about what was underneath, so it is partially recoverable.

Blur

Three stacked box-blur passes. Blurring is a mathematical operation with a partial inverse, so this is partially recoverable too.

Blur and pixelation can sometimes be partially reversed; a solid bar cannot. Choose by consequence rather than by looks — if the thing under the box would do real harm coming back, use the bar. Blur, pixelate or black box compares the three on the cases where the choice actually matters.

What it opens

JPG, PNG, WebP, GIF, BMP, AVIF and SVG, plus HEIC and HEIF as an iPhone writes them, and TIFF. HEIC and TIFF need a small decoder that is fetched only when you open one of those formats.

Camera RAW is refused, with an explanation rather than a broken preview: CR2, CR3, NEF, ARW, DNG, ORF and RAF all get turned away. Browsers cannot decode RAW reliably, and pretending otherwise would produce a mangled image you might not notice was mangled. Export a JPG or PNG from whatever you shoot with and bring that instead.

PDFs go to the document tool instead, which opens them page by page and writes the redacted file back out flattened. Still images and PDFs only — this is not a video tool. The working canvas is capped at 3,000 pixels on the long edge so a phone does not run out of memory while you drag boxes, but the file you download is always composed against the full-resolution original. Re-encoding through the canvas also drops the EXIF block as a side effect, GPS coordinates included.