Turn a photo of a page into a scan
A photograph of a document is grey, lit from one side and a degree off square. This straightens it, levels the lighting so the paper comes out white, and saves it as a PDF — optionally with the text behind it so the scan is searchable.
- Never stored
- No queue, no waiting
- No signup, no watermark
How it works
Add your photos
One page or many — they become the pages of one PDF, in the order you added them.
Watch it clean up
The skew is measured and removed, and the paper behind the ink is estimated and lifted to white, which is what makes a photograph stop looking like one.
Download the PDF
Ask for a text layer and the words are read and laid invisibly over the page, so the scan can be searched and copied from.
Turning a photograph into something that reads as a document
A photograph of a page and a scan of a page differ in three ways, and two of them are fixed here. The page is straightened — a sheet that went through at an angle, or was photographed slightly rotated, is turned level — and it is set upright, so a page captured sideways comes out the right way up. Both were necessary for the reading anyway, which is why they come free. Contrast is lifted so that paper reads as white and ink as black rather than as a grey photograph of a document.
The third difference is not fixed, and it is the one to plan around. Perspective is not corrected: a page photographed from an angle keeps its trapezoid shape, narrower at the far edge, and only its rotation is straightened. A page with a heavy fold, or one curving into the spine of a thick book, still curves. Both make the text harder to read as well as uglier — so the single most useful thing you can do is hold the camera directly above the page rather than leaning over it.
Asking for a text layer gives you a PDF that looks exactly like the photograph and behaves like a document: the recognised words are placed invisibly over the words in the picture, so the file can be searched, selected and copied while still being an image of the original. Nothing is re-typeset, and the appearance is unchanged — which is what you want for a receipt, a contract or anything where the original's look is part of the record.
Accuracy follows the capture, and the numbers are not subtle. Flat, evenly lit, in focus and square to the camera reads well; a shadow from your own head across the page, a phone flash bouncing off glossy paper, or a crumpled receipt does not. Chinese, Japanese and Korean words are recognised but counted rather than written into the text layer, because no font capable of carrying them is bundled — you are told how many, rather than left with blanks you might not notice.
When you want something else
Your phone almost certainly has this built in and better. The Notes app on iOS and Google Drive on Android both scan documents with live perspective correction — they find the page edges as you hold the camera and flatten the trapezoid properly, which is precisely the step missing here. For photographing a page, use one of those and bring the result somewhere else if you need to work on it.
For a stack of pages, OCRmyPDF is the right tool and is free: ocrmypdf --deskew --rotate-pages --clean in.pdf out.pdf does straightening, rotation, cleanup and the text layer in one pass over a document of any length, and --output-type pdfa makes it archival at the same time. unpaper handles the harder scan cleanup — removing the dark border from a book scan, splitting two-page spreads — that nothing here attempts.
Frequently asked questions
What does it actually change?
Three things. The skew: the angle of the text is measured across the whole page and removed, so the lines are level. The lighting: the brightness of the paper behind the ink is estimated in patches and lifted to a uniform white, which removes the fall-off towards one edge and the shadow of your hand. And the contrast: the ink is darkened against that white. What it does not do is invent anything — nothing is sharpened into existence.
Can it make the scan searchable?
Yes. Switch on the text layer and every word is read and placed invisibly over the picture where it appears, so Ctrl+F finds it and you can select and copy from the PDF. The page still looks exactly like the photograph; the text is never painted, only made selectable.
Is my photo stored anywhere?
No. The cleaning and the reading both happen here, and nothing is stored. Nothing is sent anywhere.
Should I crop to the page first?
It helps. The paper estimate works best when most of the frame is paper, so a photo with a dark desk around the edges comes out better if you crop it down first. The skew measurement deliberately ignores the outer margins, so a little background is not a problem.
Will it fix a photo taken at an angle?
It removes rotation — a page a few degrees off square is set level. It does not correct perspective, so a page photographed from the side, with one edge longer than the other, stays that shape. Shooting from directly above is worth the extra second.
Colour or black and white?
Colour is kept, with the paper whitened behind it — so a stamp, a signature in blue ink or a highlighted line stays the colour it was. That is usually what you want from a document; it is not a black-and-white fax.
How big will the PDF be?
Each page is stored as a JPEG sized for reading rather than at the camera's full resolution, so a phone photo that was 4 MB typically lands under 1 MB a page with no visible loss at 100%.
Good to know: Perspective is not corrected: a page photographed from an angle keeps its trapezoid shape, and only the rotation is fixed. A page with a heavy fold or a strong curve near the spine will still curve. The text layer, where you ask for one, carries the same caveats as any OCR — and Chinese, Japanese and Korean words are counted rather than written, because no font for them is bundled.
Put this tool on your website
Free for any blog, class page or help article. Paste one snippet and your visitors can use it right on your page.