Photo Censor

about this tool

Hide part of a picture

Open a photo or a screenshot — drop it in, choose it, or paste it straight onto the page — and drag across anything that should not be seen: a face, a card number, a name in a chat, a number plate, the address on a parcel. Each area you draw is hidden one of three ways. A box paints it one solid colour, black or white. Pixelate turns it into square tiles, each the average colour of the pixels it covers, spread evenly so no sliver at the edge keeps its detail. Blur spreads it out with a Gaussian blur, by the three-box approximation the W3C Filter Effects specification gives for browsers. Drag an area to move it; select it and drag its bottom-right corner to resize it, or type its position and size in pixels. Add area puts one in the middle, and in the list of areas the arrow keys move one and Delete removes it, so it all works from the keyboard.

Each effect writes only inside its own area, and no pixel from outside the areas is mixed in: the blur holds its edges rather than reaching past them, which also makes it a little weaker at the very edge, so draw a blur a little larger than what it covers. Where soft areas overlap, their effects stack. Boxes are painted before and after every blur and mosaic, so a blur that overlaps a box sees only the box's colour — it cannot smear what the box hides out past the box's edge — and the box stays solid on top.

Which effect to use

For text, use a box. Blur and pixelation both keep a trace of what was there, and they have been undone. Researchers at UC San Diego (Hill and others, 2016) trained a model on how each letter looks after the same mosaic or blur, then decoded the likeliest text from redacted documents; the open-source Depix tool (2020) recovers pixelated passwords from screenshots by matching the tiles against pixelated sample text. Both rely on knowing the effect and the font, and short, predictable text — a card number, a PIN, a one-time code — is the easiest to recover. A box replaces every pixel, so there is nothing to work back from.

Pixelation and blur are the right choice where the aim is to stop casual recognition and the picture should still look like a photo: faces in a crowd, children, number plates, a house number. Each area has its own strength, in the picture's own pixels — the tile size for pixelation, and twice the blur's standard deviation. New areas start at about a sixtieth of the picture's longer side (8 px at least, 200 at most), or at the last strength you chose; go higher for big text or close-ups.

What the saved file contains

Only the pixels you see. The download is a new PNG or JPEG painted afresh from the areas as they stand and encoded from the canvas, so under a box nothing of the original is anywhere in the file — no layers, no mask, nothing to peel back. If the picture cannot be painted, nothing is saved, and saving stays off until at least one area is drawn. The original's metadata stays behind with the original: the camera's EXIF block with its GPS position, serial numbers and date, and the small embedded preview some cameras write, which would otherwise still show the uncensored photo. The EXIF viewer shows what a photo carries before you share it. PNG is lossless, so it saves exactly what the canvas shows; JPEG makes a smaller file of a photo, and turns transparent areas white.

Limits

The file is a fresh copy, not the original with a patch: it is decoded the right way up by its EXIF orientation, colours are converted to the standard sRGB range (which can slightly flatten a wide-colour photo), a 16-bit PNG comes out at 8 bits, and an animated GIF, WebP or PNG keeps only its first frame. The picture is worked on at full resolution, so a very large photo can be slow, and some phones refuse pictures above roughly 16 megapixels on a canvas. A censored copy only protects you if it is the one you send: the original is untouched on your device. Nothing is uploaded.

questions

Is blurring or pixelating enough to hide text?

Not reliably. Both keep a trace of what was underneath, and it has been worked back out: researchers at UC San Diego decoded text from mosaiced and blurred documents with a model trained on how letters look after the effect, and the Depix tool recovers pixelated passwords from screenshots. The risk is highest for short, predictable things: card numbers, PINs, codes, names. For those, use a solid box, which replaces every pixel and leaves nothing to work back from. Blur and pixelation are fine for faces, plates and backgrounds where the point is to stop casual recognition.

Can someone remove the box afterwards?

Not from the file this tool saves. The download is a new, flat image made from the pixels on screen: where a box is, the file holds only box-coloured pixels, with no layer, mask or copy of the original underneath. That is the difference from drawing a shape over a PDF or in some editors, where the original stays in the file and the shape can be moved away.

What happens to the photo’s location and other metadata?

It is left behind. The saved file is encoded fresh from the canvas, so the camera’s EXIF block does not come with it: GPS position, camera and lens, serial numbers, the date — and the small preview image some cameras embed, which would otherwise still show the uncensored picture. To see what a photo carries before you share it, the EXIF viewer reads it all.

What does strength do?

It sets how coarse a pixelated or blurred area is, in the picture’s own pixels, and each area has its own. For pixelation it is the size of each square tile; each tile becomes the average colour of the pixels in it. For blur it is twice the standard deviation of the Gaussian, so the spread is about that many pixels. New areas take the last strength you set, starting at about a sixtieth of the picture’s longer side, 8 px at least, which hides a face in a typical photo; raise it for large text or close-ups.

Is my picture uploaded?

No. It is decoded, censored and encoded on a canvas in your browser, and the download is made on your device. You can paste a screenshot straight onto the page, and copy the result back to the clipboard as a PNG.