A redaction that can be undone is not a redaction
The failure mode this tool exists to avoid is famous and keeps happening: someone covers a password, an address or a face with a black rectangle in a document editor, exports it, and the rectangle turns out to be an object sitting on top of the original content. Selecting it and pressing delete gives the original back. It has leaked medical records, court documents and more than one set of credentials.
Here the redaction is written into the pixels. The region is replaced on the canvas and the canvas is what gets encoded, so the covered pixels do not exist in the output file. There is no layer, no object, no original underneath. Reopen the download and inspect it however you like. There is nothing to recover.
Pixelate is weaker than it looks
Pixelation is a reversible transformation in principle, and in practice for a small enough search space. If a pixelated region contains text from a known alphabet, an attacker can render every candidate string, pixelate each the same way, and compare. That has been demonstrated against pixelated bank details and redacted names. Faces are much harder but not immune.
So the rule is simple: for anything that must not be recovered, use solid fill. A solid block destroys the information rather than transforming it, and no amount of cleverness gets it back. Pixelate and blur are for making something unobtrusive, not for making it secret.
The strength setting is measured against each box rather than against the image, which matters more than it sounds. A fixed block size makes a small face a solid smear and leaves a large one perfectly readable at the same setting, and the large one is the case someone thinks is covered when it is not.
Metadata is the other half
Covering a face does nothing about the GPS coordinates in the file's EXIF block. A censored photo that still says where it was taken, on what camera, and at what second is not censored in any way that matters. This page warns when the file carries metadata, and the re-encode removes it from the output. Check the original with the EXIF viewer before sharing anything.
Common problems
- The preview shows the old result. Applying renders once. Move a box and apply again; the button is the commit point, and the file you save is always what you last applied.
- The output is a PNG, not a JPEG. Deliberate. A hard-edged block re-encoded as JPEG gets ringing artefacts around the edge, and at heavy compression that ringing can carry information about what was underneath.
- The blur has a visible seam. A blur samples pixels from outside its own region, so the whole image is redrawn through the filter and clipped to the box. Without that it would sample transparency at the edges and leave a dark halo.
- Someone can still tell what was there from context. Redaction does not remove the shape, position or size of what was covered. A single blacked-out word of known length in a known sentence is not private.
Frequently asked questions
Is the image uploaded?
No. Everything happens in a Web Worker in this tab. For a redaction tool that is not a nice extra. Sending an unredacted image to a server in order to redact it means the unredacted version has already left your control.
Can I redact a PDF here?
No. A PDF holds text as text, so covering it visually leaves the words extractable underneath, the exact failure described above. Redacting a PDF properly means removing the content stream, which is a different job from anything on this page.
Does it detect faces automatically?
No, and that is a deliberate limitation. Automatic detection misses faces at angles, in shadow and in reflections, and a tool that says "3 faces redacted" invites you to trust it and stop looking. Placing the boxes yourself means you have looked at the whole image.
Can I check the redaction worked?
Yes, and you should. Save the file, then reopen it here or anywhere else and zoom into the covered region. What you see is all the data there is.