Deskew Scanned PDF
Straighten a tilted scan: every page measured on its own, adjustable by hand, and pages that are already straight are left alone. Free, no upload.
Drop a scanned PDF here or click to browse
🔒 Straightened entirely in your browser — your PDF is never uploaded.
Straighten crooked scanned pages
Scanned or photographed documents almost never land perfectly square — each page sits a degree or two off, which looks untidy and trips up OCR. Drop in a PDF and this tool renders every page, measures how far its text is tilted, and rotates each page back to level on a clean white background. Detection uses a projection-profile method: it tries a range of small angles and keeps the one that makes the text lines the sharpest and most horizontal, correcting up to about 15 degrees in either direction. Each page is judged on its own, so a mixed stack straightens correctly. Because the measurement follows the lines of text, a page that is mostly photos, tables or a large signature can fool it — so the measured angle is shown in plain words and you can overrule it: set any angle within ±15° by slider or by typing it, nudge it a tenth of a degree at a time, and watch a live before/after preview of your own page before you download. A hand-set angle belongs to the page you set it on, so the other pages keep the angle measured for them; one button applies it to all of them when the whole scan came out of the feeder leaning the same way. There is also a box for straightening only some pages ("1-3, 8"), and any page that measures straighter than 0.4° is left exactly as it is rather than re-drawn, so it keeps its selectable text.
Private by design
Everything runs locally with pdf.js and pdf-lib — pages are rendered to a canvas, analysed, rotated and re-embedded into a brand-new PDF entirely in your browser, so your document is never uploaded to a server and it keeps working offline once loaded. A page that is turned is rebuilt from its rotated image, so that page becomes a picture: faithful to the original layout but with no selectable text, and very sharp corners can be trimmed slightly by the rotation. A page that is left alone — already straight, or outside the pages you chose — is copied across untouched and keeps everything it had. It is text-faithful, not pixel-perfect — a page with almost no text may be left as-is when no reliable tilt can be measured.
Frequently asked questions
How much skew can it correct?
Up to about 15 degrees of tilt in either direction. Each page is measured independently, and a page that measures straighter than 0.4° is left exactly as it was — copied across rather than re-drawn, so it keeps its selectable text. Extreme tilts beyond the search range are only partly corrected.
The automatic angle is wrong on my page. Can I fix it?
Yes. The measured tilt is shown next to a slider you can drag, type into, or nudge by a tenth of a degree, and the preview updates as you do. The angle you set belongs to the page in the "Look at page" box, so you can correct one crooked page without disturbing the measurement on the rest; press "Measure this page again" to hand that page back to the measurement, or "Use this turn on every page" when the whole scan leans the same way.
Will the text still be selectable?
Only on the pages that were actually turned. A page being straightened is rendered, rotated and rebuilt as an image, so it becomes a picture of the page with no selectable or searchable text — the same trade-off as a fresh scan. A page that was already straight, or that you left out of the run, is copied across untouched and keeps its text. Keep the original if you need the whole text layer, or run the deskewed file through OCR afterwards.
Is my PDF uploaded anywhere?
No. Rendering, skew detection and rotation all happen in your browser with pdf.js and pdf-lib, so your file never leaves your device and the tool works offline once the page has loaded.