PDFを墨消し

このツールでできること

Redaction here removes the content rather than covering it. Every page carrying a redaction is rebuilt from pixels: the boxes are drawn into the page, the page is rendered at the resolution you choose, and the render becomes the new page. The original content stream is discarded rather than edited, so there is nothing underneath to select, copy or search. Pages with no redactions are copied through untouched.

Whenever a document has to go out with something taken out of it: names in a court filing, salaries in a board pack, an account number on an invoice, patient details in a medical record, an internal comment in a contract going to the other side.

仕組み

1. Drop the PDF onto this page and draw boxes over whatever has to go. Each box is one redaction.
2. Add a phrase under find and redact to catch every occurrence of a name or a number wherever it appears, instead of hunting page by page.
3. Set the resolution for the rebuilt pages. 200 dpi reads well on screen and prints acceptably; 300 is better for print and produces a larger file.
4. Leave keep the document searchable on unless you have a reason not to, and leave remove metadata on - a document's properties often hold the very name you have just taken off the page.
5. Press Redact, then open the result and try to select the redacted text. There will be nothing there.

Drawing a black rectangle over text is not redaction, and the mistake has leaked court records, medical files and government documents repeatedly for twenty years. The rectangle is a drawing instruction added on top; the text underneath is still in the content stream, where anyone can select it, copy it or pull it out with a two-line script. The same is true of cropping.

Rebuilding from pixels is the only approach a browser can take that is genuinely safe, and it is deliberately blunt. The boxes are drawn into the document first and the page is rendered afterwards rather than the other way round, so no rounding in raster space can leave a one-pixel sliver of the original visible along an edge. There is no underneath, because the underneath was thrown away.

Losing search would be a real cost, so the surviving text is put back. The text layer is extracted before anything is drawn, every run falling inside a redaction box is dropped, and the rest is drawn back invisibly over the image. The document stays searchable and the redacted words are genuinely absent - the combination people assume a black box gives them and it almost never does.

The trade-off is size and fidelity on the pages you touched. A vector page that becomes a 200 dpi JPEG is usually several times larger, and links, form fields and annotations on that page do not survive the rebuild - which is why only affected pages are rebuilt. Check the result rather than trusting it: search for the word you removed, then select across the box and paste.