Straighten scan orientation

Vellum works out the reading direction of each scanned page: the way text lines are spread tells first whether the page lies on its side or upright, then the way ink is distributed inside those lines tells whether it is the right way up or upside down; only when that is inconclusive does text recognition decide between the two remaining options. The proposed rotation appears on each thumbnail, already applied to the preview, and one click turns the page a further quarter turn. The correction is written into the PDF as a display instruction: nothing is rasterised, no quality is lost, and the change stays reversible. A page without text — a photo, an illustration, a blank sheet — offers no lines to measure and is left untouched. Recognition models and rendering are served from your own device: the document is never transmitted.

How does Vellum work out which way up a scanned page is?

In two steps. It first measures how ink is spread across the page: horizontal text alternates inked lines with empty gaps, which separates upright pages from pages lying on their side without reading anything. To then tell an upright page from an upside-down one, text recognition reads a strip in the middle of the page both ways and keeps the direction it is most confident about — the only method that proved reliable in testing. When recognition is not available, Vellum corrects the axis only and leaves the page as it is rather than guess.

Does rotating degrade the quality of the scan?

No. The rotation is written into the PDF as a display instruction: page content is not turned pixel by pixel, rasterised or recompressed. The original sharpness is preserved, text stays selectable, and the change can be undone by applying the opposite rotation.

Why are some pages left unchanged?

Because they offer nothing to measure. A full-page photo, an illustration, a flat colour area or a blank sheet has no text lines: Vellum says so on the thumbnail and leaves the page alone rather than guessing. You can still turn it yourself by clicking on it.

Does the text recognition download anything?

No. The recognition engine and its language model are served from your own device, as they are for Vellum’s OCR tool, and they are loaded only if a page genuinely needs them. On a document whose text lines settle the question — the usual case — recognition never starts.

Vellum — free PDF tools whose processing stays on your device.