Can the covered area be recovered from the exported image?

Not if you used a solid block. The pixels are replaced and the file is rebuilt from pixel data, so there is no layer to remove and nothing of the original left in those areas. The check that runs after every export confirms this against the exported file rather than assuming it. Pixelation and blur are different — both leave the information in a reduced form and both are recoverable.

Why is pixelation not good enough?

Because it is a deterministic transformation of data that is still there. If someone can guess what sort of thing was written — a name, a number, a postcode — they can pixelate every candidate the same way and compare until one matches. This has been done to real published documents, and text is the easiest case because the character set is small.

Does it remove the photo's metadata?

Yes, as a consequence of how the export works. The image is rebuilt from pixels, so nothing that is not pixel data survives: no GPS coordinates, no camera serial number, no original filename, and no embedded thumbnail. That last one matters — a JPEG often contains a small copy of the picture as it was before any editing.

Is my image uploaded?

No. The file is read, edited and exported entirely in your browser. Nothing is sent anywhere, and the verification also runs locally.

What does the check actually compare?

The exported file is decoded again and compared with the original pixel by pixel. It reports how many pixels inside your covered areas changed, whether those areas are now a single flat colour, and whether anything outside them changed. Checking the exported file rather than the editing canvas is the point: it confirms what a recipient would receive.

Which format should I choose?

PNG unless the file size is a problem. It does not re-encode, so a hard-edged block stays hard-edged and nothing outside your covered areas changes. JPEG is smaller but re-encodes the whole picture and can leave visible artefacts around the block edges.

Why is the default filename not based on the original?

Because a file called passport-redacted.png published next to passport.png tells anyone looking what they are looking at, and the original name often names the document. You can keep the original name if you want it.

Does this work on PDFs?

No — this covers areas of an image. For a PDF, use the PDF redaction tool, which flattens the page so the text underneath cannot be selected or recovered.

Image Redactor

Cover something in a picture so that it is actually gone — then check, pixel by pixel, that it is. The exported image is flattened, its metadata is dropped, and the result is read back and compared against the original before anything is claimed.

Choose an image

or drop one here — it never leaves your device

Three ways a redaction is not one

The box is a separate layer. Draw a black rectangle in a PDF or a design tool and the original is still underneath. Selecting the text, deleting the shape or opening the file in a different program brings it straight back. This has caused a long series of real disclosures, including court filings and government documents. The only reliable answer is to flatten to pixels, which is what happens here: the export is a new image built from pixel data, with no layers to peel back.

The box is solid but the file still tells you everything. A photograph carries the GPS coordinates where it was taken, the camera and often its serial number, the original filename, the time to the second, and frequently an embedded thumbnail of the picture as it was before any editing. Covering a face achieves nothing if the file contains a small copy of the same picture uncovered. Exporting through a canvas drops all of it, because none of it is pixel data.

The area is blurred or pixelated rather than covered. Both reduce the information. Neither removes it.

Why pixelation is not redaction

This deserves stating plainly, because pixelation looks thorough and is the most common choice.

Pixelating a region averages it into blocks. That is a deterministic transformation: the same input always produces the same output. So if an attacker can guess the sort of thing that was there — a name, a card number, a postcode, a word from a small vocabulary — they can pixelate every candidate the same way and compare the results until one matches. There is nothing to break; it is a lookup.

This is not theoretical. It has been done to published documents and to broadcast footage, and tooling for it is freely available. Text is the easiest case of all, because the character set is tiny and the layout is regular.

Blur is the same story with more mathematics. Even a heavy blur leaves the length, rhythm and rough shape of what was there, and a light one is close to reversible.

Use either to show that something has been removed. Never use them to hide what it said.

The check that runs after every export

Everything above is a description of intent. The verification below is a measurement.

After the redaction is applied, the exported image is decoded again and compared with the original, pixel by pixel. Three things are reported: how many pixels inside your covered areas actually changed, whether those areas are now a single flat colour — which only a solid block can achieve — and whether anything outside them changed, which would mean the whole picture was re-encoded.

It is a small check, and it is the difference between being told the redaction worked and being shown it.

Before you publish it

Redact the largest sensible area rather than the smallest. A box tight around a name leaves its length and shape, and on a form the position of a field can be as identifying as its contents.

Think about what is left. A redacted street name next to an unredacted house number and an unredacted photograph of the front door is not redacted. The surrounding detail is usually what identifies someone, not the piece you covered.

And check the filename. passport-redacted.png sitting next to passport.png tells anyone looking exactly what they are looking at, and the original name often names the document. The default here is a neutral name for that reason.