Skip to content

How to extract text from an image or scanned PDF

OCR without uploading your documents: what reads perfectly, what needs a re-shoot, and the scanned-PDF flow.

Updated September 2, 2026 · 5 min read · by the InfyTool team

A phone photo of a form, a screenshot of an error, a scanned contract — the text is right there, and yet you're about to retype it. OCR (optical character recognition) reads it out in seconds, and the version here runs entirely on your device, which matters because documents worth OCR-ing are usually documents worth keeping private.

Images and screenshots

  1. Open Image to Text and drop the photos — several at once is fine.
  2. Pick the language of the text (the model downloads once, then works offline).
  3. Each image is read with a confidence score; copy per image, or download everything as one .txt.
Extract text from an image now

Runs on-device — IDs and documents never touch a server.

Scanned PDFs

Drop the file into PDF to Text. A PDF made from Word or a website extracts instantly (it carries real text); a scanner's PDF is photographs of pages, and the tool detects that and offers Run OCR — each page renders at 300 DPI and is read in sequence. A 20-page scan takes a couple of minutes and never leaves your machine.

Getting the best read

OCR loves flat, sharp, well-lit printed text. Photograph the page straight-on and fill the frame; screenshots read near-perfectly as-is. Watch the confidence score — below ~80% usually means a blurry or angled shot, and re-shooting beats hand-correcting. For text living in a Word file there's no OCR needed at all: the Word converter extracts it directly, and EPUB to Text does the same for ebooks.

Frequently asked questions

How accurate is browser OCR?

On clear printed text — documents, books, screenshots, signage — very accurate, and every result shows a confidence score so you know when to trust it. Handwriting is honestly out of scope: that requires cloud-scale models no on-device tool provides.

Which languages are supported?

English, Hindi, Spanish, French and German — each a compact model (1–3 MB) that downloads once when selected and then works offline.

Can it read a scanned PDF?

Yes — drop the scan into PDF to Text: when no text layer is found, it offers on-device OCR that renders each page at 300 DPI and reads the whole document, with per-page progress.

Why does no-upload matter for OCR in particular?

Because of what people OCR: IDs, bills, contracts, prescriptions, salary slips. Every other OCR site ships those images to a server; here the Tesseract engine runs in your browser, and you can go offline and keep extracting.

Tools used in this guide