Remove blank PDF pages

Blankness is scored from pixels first and content signals second. Every page is rendered to a canvas of roughly 100 by 140 pixels, and each pixel whose red, green and blue values are not all at least 250 counts as ink; the ratio of inked pixels to total pixels is that page's score. Compare the score to the sensitivity threshold and you have a first verdict. On Normal a second test can veto removal: if the page still reports selectable text once whitespace and zero-width characters are stripped, or a visible annotation that is not a link or a popup, or a vector fill, stroke or shading in a colour other than near-white, the page is kept however empty it looks.

Strict, Normal and Loose set the ink threshold, and Normal alone adds the content veto. Strict flags a page only below 0.1% inked pixels, Normal below 0.5%, Loose below 2% — so Loose will also catch a page carrying a lone footer or a faint scan edge. Strict and Loose judge pixels and nothing else, which is deliberately what scanned documents need, because a scan has no selectable text for the veto to find. Normal is the default because it will not silently delete a page that still carries real characters, an annotation or a drawn line. Switching between the three is instant: the scores already exist, so only the classification is recomputed.

On Normal, a page number is content, and content vetoes automatic removal. The text test does not care how much text there is; one visible character is enough for the page to be kept. That is intentional, since deleting a page that carried a section heading or a numbered footer is a far worse outcome than leaving one nearly empty page in the file. There are two ways round it: switch to Loose, which scores pixels only and will flag a page whose ink covers under 2% of its area, or click the thumbnail and flag it yourself. Manual selection always overrides whatever the detector concluded. Either way the strip updates immediately, so you can see the new verdict before you save anything.

False positives are possible, most often on scans, which is exactly why nothing is removed until you have looked. A scanned page is one big image, so faint handwriting or pale pencil marks are pixels rather than characters and the text veto has nothing to grab; if those marks cover less than the threshold, the page gets flagged. Very light watermarks and pale grey backgrounds behave the same way. The workflow is built around that: every page appears as a thumbnail with its verdict, clicking one flips it, and the action bar reports exactly how many pages will disappear. Pages that could not be checked at all get an amber note and are kept by default.

No — the trimmed file is a new document, so document-level structures do not survive. The pages you keep are copied into a freshly created PDF in their original order, which is what guarantees a clean file with no leftovers from the pages you removed, but it also means the Info dictionary is reset and any outline entries pointing at deleted pages are gone. Restore the descriptive fields afterwards with Edit PDF metadata, and rebuild an outline with Bookmarks editor if the document had one. Page content itself is copied faithfully, so text stays searchable and images keep their original quality. On a bulk scan, which rarely carries an outline or meaningful document properties in the first place, nothing of value is lost.

Removing blank pages is free, unlimited and watermark-free, and the scan happens entirely on your own machine. Detection uses pdf.js to rasterize pages in your browser and pdf-lib to write the trimmed copy; no page image, thumbnail or score is ever sent anywhere. That is worth knowing here, because the documents people de-blank tend to be bulk scans of contracts, invoices and correspondence. The cost of local processing is time: a several-hundred-page scan is rendered page by page and takes noticeably longer on a phone than on a laptop. Encrypted PDFs need unlocking with Remove PDF password first. Closing the tab discards the thumbnails and the scores along with everything else, and reopening the page starts a completely fresh scan.