Blur or pixelate part of a photo

Covered, or actually gone?

There is a difference between hiding something visually and removing it from the file, and it is the difference that gets people into trouble. Drawing a black rectangle in a PDF, a Word document or a layered image editor leaves the original underneath: the shape sits on top, and anyone can move it, delete the layer, or read the text straight out of the file. Redactions have been undone that way in court filings and government releases more than once.

That cannot happen here. What you get back is a flat re-encode of the finished picture — no layers, no history, no original underneath. The covered pixels are not in the file because they were never written to it.

Which effect actually hides something

A solid block is the only one that hides everything. It writes one colour over the area, so what was underneath contributes nothing at all to the output. If you are covering something that must not come back — a password, an account number, a home address — use this and stop reading.

Pixelation and blur keep an averaged trace of what they cover, and that trace can be attacked. When the hidden content is drawn from a small, guessable set — digits, a short word, a number plate against a known font — someone can render every candidate, pixelate each one the same way, and see which matches. This is a real, demonstrated technique, not a theoretical worry. Against a face in a crowd it is impractical; against a six-digit number it is not.

Both effects here work the same way: the area is scaled down to a handful of pixels and scaled back up, with smoothing off for a mosaic and on for a blur. That is a deliberate choice over a convolution blur filter. Resampling throws detail away; a filter rearranges detail that is all still present. What comes out of this tool cannot be sharpened back, because there is nothing left to sharpen — but it can still be guessed at, which is why the paragraph above matters.

The strength setting is expressed as blocks across the short edge rather than a pixel size, so "strong" means the same thing on a thumbnail and on a 48-megapixel photo. Four blocks across a face is unrecognisable; sixteen is often not.

What else the file gives away

Covering a face does not cover the fact that the photo carries GPS coordinates. Phone cameras write EXIF metadata into every shot — the exact location, the time, the camera and often its serial number — and that survives being cropped, resized and posted almost everywhere.

It does not survive this tool. The output is written from decoded pixels, so no metadata block is carried across. If that is the only thing you needed, you can save a copy without covering anything at all; the tool will tell you that is what happened.

Two things it cannot help with. Screenshots often show more than you meant in the margins — notification banners, browser tabs, the other half of a conversation. And a reflection in a window or a pair of glasses has given away more locations than metadata ever has. Look at the whole frame before you post it, not just the part you came to cover.

Questions

Can I cover more than one thing?

Yes, and each area keeps the settings it was drawn with. Choose a solid block, drag over the account number, then switch to pixelate and drag over the faces — both are applied when you save. Every area is listed underneath the photo and can be removed individually.

Does saving reduce the quality of the rest of the photo?

It is re-encoded, so a JPG goes through one more generation of lossy compression. At the default quality that is invisible in practice. Choose PNG if you want the untouched parts bit-for-bit intact, at the cost of a much larger file.

Can I undo a box after I have drawn it?

Yes — undo the last one, remove any one from the list, or clear them all. Nothing is committed until you save, and the photo on your disk is never modified either way.

Does this work on a phone?

Yes. Dragging on the canvas draws a box rather than scrolling the page. On a small screen it is worth zooming the browser in before drawing a tight box around something small.

Can it find faces automatically?

No, and that is on purpose. Automatic detection needs a model of tens of megabytes, which would slow the page down for everyone, and it misses faces — turned away, partly hidden, at the edge of the frame. A miss you do not notice is worse than a box you have to drag yourself.

What about video?

Not here. Censoring video means tracking the region across every frame and re-encoding the whole file, which is a different tool and a much heavier one.

I need the file smaller before I can send it.

Save the censored copy first, then put it throughcompress to a target size orthe resizer. Do it in that order — compressing first and censoring second just re-encodes twice.