Vellum computes a visual fingerprint for every page — a 64-bit signature derived from the low frequencies of the image, unaffected by scanner noise or a few millimetres of offset — then groups the pages whose fingerprints match. When the document has a text layer, that text is compared too, so two copies of the same form filled in with different values are not mistaken for one another. Groups are shown side by side: the first occurrence is kept, the later ones are proposed for removal and can be unticked. The resulting file keeps the original pages without recompressing them, and the whole analysis runs in your browser.
It computes a perceptual fingerprint: the page is shrunk to a thumbnail, converted into frequencies, and only the lowest of them — the overall structure of the page, not its detail — produce a 64-bit signature. Two pages are grouped when their signatures differ by only a few bits. This shrugs off scanner noise, a slight offset or a different contrast setting, where a byte-for-byte comparison would fail.
That is the limitation to know about. Two copies of the same form look almost pixel for pixel alike; only a few handwritten or typed values set them apart. When the PDF has a text layer, Vellum compares that too and refuses to group two pages whose text differs. On a scan that was never run through OCR, that safeguard does not exist: review the thumbnails before confirming the removal.
The first occurrence, in document order. The later ones are shown ticked and ready to be removed, and any tick box can be cleared with a click if you would rather keep two copies. The tolerance slider also lets you tighten or loosen what counts as “identical pages”.
Yes. Computing the fingerprints, grouping the pages and building the final PDF all happen in your browser, with no page sent to a server and no account to create.
Vellum — free PDF tools whose processing stays on your device.