Censor an image so the covered part is gone from the file

Because a black rectangle in a PDF is a drawing instruction and the text underneath it is still a separate object in the same file. The viewer paints the rectangle last, so it looks covered; a copy-and-paste, a text extractor or a different viewer's rendering order gets the words. Several published documents have come apart exactly this way, and the pattern is always the same — the covering and the covered content were two things in one file, and the reader received both. Turning the page into pixels first breaks that, because a bitmap has nowhere to keep the text.

Only if it was flattened into the pixels before the file was written. An editor that keeps the rectangle on its own layer — a PSD, an XCF, a TIFF with layers, or an editor's own project format — is describing the same failure the PDF has, and so is any export that carries an embedded original alongside the edited version. Flatten, export to a format with no layer support, and reopen the exported file to look at it. If the shape can still be selected and moved, nothing has been removed yet.

It has, on a large scale, and the cause was a file being overwritten rather than rewritten. Screenshot editors on two major platforms wrote the shorter cropped image over the longer original without truncating the file, leaving the tail of the old image sitting past the end of the new one where it could be recovered — the bugs were published in March 2023 as CVE-2023-21036 and CVE-2023-28303. It is worth knowing here because it is the same mistake as the layered redaction wearing different clothes: the new content was placed on top of the old rather than in place of it. Every export from this page is a fresh encode of a fresh canvas, so there is no older, longer version for anything to hide in.

No. The values under each area are overwritten in a canvas and the file is encoded from that canvas afterwards, so there is no second copy for them to survive in. A PNG written this way has one image stream, no layers, no embedded preview and no thumbnail — the small picture your operating system shows in a folder is generated by the operating system from the image itself, not stored inside it. The panel above the download button states the four facts the file can be checked against: one bitmap, zero layers, the exact count of pixels overwritten, and zero metadata records.

Use the bar for anything a person could read or a machine could match, which means text, numbers, barcodes, faces you want gone rather than softened, and signatures. Blur and blocks both keep a low-resolution version of what was there — that is what they are, mathematically — and low-resolution versions of short strings are recoverable by testing candidates against them. The soft fills earn their place where the goal is discretion rather than deletion: a face in a background, a poster on a wall, a screen at the far end of a room.

Because redacting one in the browser would reproduce the bug this page exists to avoid. This tool works on pixels, and pixels are the only form in which a covered word is genuinely gone; a document format keeps text as text, and any covering placed in it is a shape drawn over content that is still there. Export or print the page to an image first — every operating system and every PDF viewer will do it — and redact the image. The resulting file is not searchable any more, which is the trade being made and usually the point.

Not after this page. A camera writes EXIF and often GPS coordinates into the original, and those tags survive most editing tools untouched — a photograph of a document with a redacted address can still carry the coordinates of the room it was photographed in. Decoding to raw pixels and re-encoding through a canvas drops every one of them, along with the camera make, the serial number, the lens, the capture timestamp and any XMP block. That is a side effect of how the export is built rather than a feature bolted on, which is why it cannot be forgotten.