Compare PDF

How it works

Open the new version and add the old one in the panel itself (or pick an earlier version made in this session). You get a summary with the changed pages and the added and removed words, the list of pages and, for each one, the text with what was added underlined in green and what was removed struck through in red. The next and previous change buttons take you from one change to the next; you can ignore differences in spacing or in upper and lower case.

The visual changes view draws both pages and marks in red the pixels that differ, with three ways to look at them (differences, side by side and a slider) and a tolerance you can adjust. It also works for images and for scanned pages. If a page has no text, you can run OCR on your device to compare its words too.

You can download a PDF report with only the pages that changed, or the list of changes as a text file. Neither download changes the open document, and both files are read and compared in your device's memory.

What is not preserved, and limits

• Only the text layer of each page is compared (and, separately, the pixels). Text that is part of an image is not read unless you run OCR.
• Moved or reordered content shows as removed in one place and added in another: the tool does not know it is the same paragraph. It only gives a hint when identical text seems to have moved.
• Ligatures (such as fi), no-break spaces and invisible characters are normalized, so they do not count as changes; even so, the fonts and encodings of some files can produce spurious differences or hide real ones. Check what matters in the visual view.
• Scanned pages have no text to compare until OCR is run (it is offered when available and never runs by itself), and the resulting text is only as reliable as the OCR; it can contain recognition mistakes. The OCR runs on an in-memory copy: your files are not touched.
• Pages are matched by page number, not by content. If a page was inserted in the new version, every later page is compared with the wrong page and shows as different. For now there is no way to offset the matching.
• Pixel differences depend on the program that draws the pages and on the tolerance chosen: the same drawing can be anti-aliased differently in another viewer, and a high tolerance can hide faint differences.
• If the two pages have different sizes, they are anchored at the top-left corner and the extra area counts as changed, or the new page is fitted to the old size, which is only approximate. A rotation difference is stated and what a viewer shows is compared. Very large pages are drawn at reduced detail and may miss small differences.
• On a page with extreme differences the calculation is cut short and the rest is shown as one removed block and one added block.
• The PDF report contains only the pages that changed, with page images at reduced resolution (compressed) and a cap on how many pages carry an image; characters that the bundled font (Noto Sans, Latin set) does not have appear as ? and are listed. The report uses a single font, with no bold.
• The result has been checked with pikepdf and qpdf (structure), poppler (pdftotext, pdftoppm) and word and pixel counts made separately; it has not been tried in Adobe Acrobat or Reader, and has only been seen working in Chromium.
• Everything happens in your device's memory and nothing leaves it. Very large files (more than about 100-150 MB between the two) can fail, especially on phones. Closing or changing the document discards the comparison.