Document scanner

Scan, clean & make one searchable PDF

Put document photos in order, optionally remove shadows and page curl, then download a locally assembled PDF with selectable text.

Processed locally and securely on our servers
  1. 1. Pages
  2. 2. Clean
  3. 3. Searchable PDF
Drop document pages
JPG · PNG · WebP · paste, folder, or camera — ordered into one PDF
or press Ctrl+V
PNG · JPG · WEBP · max 40.0 MB

Pages stay in drop order. Use the arrows to change the PDF order; paste works anywhere on this step.

How to scan a document to a searchable PDF

  1. Photograph the pages

    Lay each page flat on a contrasting surface, fill the frame with it, hold the camera parallel rather than at an angle, and light it from the side so your own shadow does not fall across the text. Thirty seconds of care here is worth more than any setting further down.

  2. Add the pages and put them in order

    Drop up to 50 images onto the page and drag them into reading order. Ordering happens entirely in your browser, so nothing is uploaded at this stage.

  3. Clean the pages, or skip it

    Switch on any combination of shadow removal, dewarping, contrast enhancement and 2× upscaling if the pages need it. Cleanup runs on a GPU and costs 2 credits per page. Skip it for a document that must not leave your device, or when the photos are already clean.

  4. Choose OCR languages

    Pick the language or languages the document is written in. Recognition runs locally with Tesseract, and each language's data file downloads on first use, so the first run in a new language takes a little longer.

  5. Build and download the PDF

    Set a target file size if the document has to fit an upload limit, then build. OCR and assembly run in your browser and the finished PDF — images plus an invisible, selectable text layer — downloads straight to your device.

The complete guide to scanning documents into a searchable PDF

Why a folder of photos is not a document

Photographing paperwork with a phone is now the normal way to digitise it, and for good reason: the camera in your pocket out-resolves most flatbed scanners and is always with you. What it produces, though, is a folder of JPEGs. That is fine for keeping a personal copy and unworkable for anything else. You cannot email nine separate images as a contract. You cannot upload a scattered set of files to a portal expecting one document. And you cannot search a single word of any of it.

A PDF solves the container problem: one file, in order, that opens the same way everywhere and prints with predictable page boundaries. But a PDF built naively from photographs is still only a stack of pictures. It looks like a document to a human and remains completely opaque to every system that has to process it.

The missing piece is the text layer, and it is what separates a scan you can use from one you merely have. This tool does both jobs in one pass: it assembles ordered pages into a single PDF, and it runs optical character recognition over each one to embed an invisible, precisely positioned text layer underneath the image. The result looks exactly like your photographs and behaves like a text document.

What OCR gives you that a photo cannot

Optical character recognition reads the shapes in an image and works out which characters they are. In a searchable PDF the recognised words are written back into the file as invisible text, positioned over the pixels they came from. Selecting a paragraph selects real text. Searching the document finds real matches. Copying pulls out characters rather than a screenshot.

That changes what the file is capable of downstream. Full-text search across a document archive starts working. Screen readers can read the page aloud, which for a scanned document is the difference between accessible and useless. Compliance and e-discovery systems can index the contents. And you can pull a reference number out of an invoice without retyping it from the screen.

Accuracy depends on the input, and the relationship is unforgiving in one direction: recognition is close to perfect on clean printed text and unreliable on anything that is blurred, skewed, low-resolution or unevenly lit. It is worth understanding this before you photograph twenty pages, because the fix is almost always at capture time. Fill the frame with the page, hold the camera parallel to it, and photograph in even light rather than under a single overhead lamp.

The cleanup pass, and when it pays for itself

Real-world document photographs have four characteristic problems, and the optional cleanup pass addresses each of them separately so you only pay for what you need. Shadows are the most common: your own hand, a lamp or a window produces a gradient across the page that makes half of it darker than the other half, which defeats any simple threshold and confuses recognition. The de-shadow stage flattens that gradient so the page is evenly lit.

Page curl is next, and it is the reason photographs of bound books read worse than loose sheets. Text near the spine curves away from the camera, so the lines bend and the characters compress. Dewarping flattens that curvature back to a rectangle. Contrast enhancement handles faint originals — pencil, faded carbon copies, thermal receipts that have started to disappear — by pushing grey paper toward white and grey ink toward black. And the 2× upscale doubles resolution, which is what small text photographed from too far away actually needs.

Each stage is a switch, and all of them together still cost 2 credits per page, so there is no reason to be sparing with the combination once you have decided a page needs help. The judgement worth making is whether it needs help at all. Photograph one page well, build a PDF without cleanup, and look at the text layer. If it is accurate, save your credits for the pages that are genuinely difficult.

Photographing pages so they scan well

The cheapest quality improvement is not a setting; it is thirty seconds of care at capture. Put the page on a flat, contrasting surface — a dark desk under white paper — so its edges are unambiguous. Fill the frame with the page and hold the phone parallel to it rather than at an angle, because perspective distortion turns a rectangle into a trapezoid and every line of text into a slightly different size.

Light it from the side or from two directions rather than from directly overhead, which is what puts your own shadow on the page. Avoid direct flash on glossy paper, where the specular highlight will erase a paragraph completely. And tap to focus before you shoot: phone cameras focus on whatever is closest, which at close range is often the edge of the desk rather than the text.

Keep the pages in order as you shoot, so you are not reordering nine near-identical thumbnails afterwards. The tool lets you reorder freely, but recognising which photograph is page six of a contract from a thumbnail is genuinely hard, and it is much easier to avoid the problem than to solve it.

What stays on your device, and what does not

This is a hybrid tool and the split is worth being precise about. Adding pages, reordering them, running OCR and building the PDF all happen in your browser. No file leaves your device for any of it, and there is no server-side copy, which is why those steps are free and unlimited — there is no infrastructure for us to pay for.

The AI cleanup pass is the exception. Shadow removal, dewarping, contrast enhancement and upscaling need a GPU, so pages you send through cleanup are uploaded, processed and returned. The page carries a hybrid-processing badge to make that visible rather than leaving you to infer it, and it is the only step that costs credits.

The practical consequence is that you have a genuine choice. For a document that must not leave your machine — a signed contract, medical records, anything covered by a confidentiality obligation — skip cleanup and the whole workflow stays local. For a difficult scan where accuracy matters more, the cleanup pass is available at 2 credits per page and the trade-off is yours to make rather than ours.

Frequently asked questions

How do I turn phone photos into a single PDF?

Add the photos, put them in page order, optionally run the AI cleanup pass, then build the PDF. Every page is combined into one document with a selectable text layer, and the file downloads straight to your device. Ordering, OCR and PDF assembly all run in your browser and are free.

What does a searchable PDF mean?

It means the PDF contains an invisible text layer aligned with the words in the image, so you can select, copy and search the text even though each page is a photograph. Without that layer a scan is just a picture of a document: it looks readable, but no search, no screen reader and no indexing system can get at a single word of it.

Is this free?

OCR and PDF assembly are free, unlimited and local — they never leave your browser, so there is nothing for us to charge for. The optional AI cleanup pass runs on a GPU and costs 2 credits per page, which is a Pro feature. You can skip cleanup entirely and still get a searchable PDF for nothing.

How many pages can I scan at once?

Up to 50 pages in one document. That covers the overwhelming majority of contracts, reports, receipts batches and lecture notes. For anything longer, build it in parts and merge the PDFs afterwards — a page count in the hundreds would exhaust browser memory before it exhausted your patience.

What does the cleanup pass actually do?

Four stages, each independently switchable. Remove shadows evens out the gradient a hand or a lamp casts across the page. Dewarp flattens the curl of a bound book or a page that will not lie flat. Enhance contrast pushes grey paper toward white and grey ink toward black. Upscale 2× doubles the resolution, which matters when the text is small relative to the frame. You pay 2 credits per page whichever combination you choose.

Do I need the cleanup pass?

Not always. A page photographed flat, in even light, filling most of the frame will OCR well without any help. Cleanup earns its cost on the awkward cases: a shadow across half the page, a curved page in a bound book, faint pencil or thermal-receipt printing, and photographs taken from far enough back that the text is only a few dozen pixels tall. Try one page without it first and see what the text layer looks like.

Which languages does the OCR support?

You can pick one or several from the language list, with English selected by default. Recognition runs on Tesseract in your browser, and the data file for each language you select is downloaded on first use, so the first run in a new language takes a little longer than the ones after it. Selecting several languages at once is the right approach for genuinely mixed documents and the wrong one for monolingual pages, where it only adds confusion.

Are my documents uploaded?

Only if you enable AI cleanup. Page ordering, OCR and PDF assembly all happen on your device, so a scan built without cleanup never touches a server at all. The cleanup pass needs a GPU, so pages sent through it are uploaded, processed and returned — the page marks this with a hybrid-processing badge rather than hiding it. For documents that must not leave your machine, skip cleanup.

How do I control the size of the finished PDF?

There is a target-size control on the final step: keep the original images for best quality, or aim for roughly 2 MB or 5 MB in total. The target is shared across all your pages, and PDF structure adds a little overhead, so the final file can land slightly above the number you picked. For emailing or uploading to a portal with a size cap, aim one step below the limit.

Can I edit the PDF afterwards?

The PDF this tool produces is a finished document — images plus a text layer — rather than an editable one. If you need to reorder pages, sign it, fill a form or merge it with another file, download it here and open it in a PDF editor. There is a direct link to OpusPDF on the result screen for exactly that hand-off.