Pull the text out of a scan or a photo of a document
Someone sent a photographed contract, or a PDF that is really just pictures of pages. You need the words, not a picture of the words. This reads it on your own machine and hands you plain text.
- Engine
- Tesseract 5 compiled to WebAssembly, about 4 MB with one language
- Reads
- PDF, JPG, PNG, WEBP, and iPhone HEIC
- Languages
- 25, including Arabic, Chinese, Hindi and Yoruba
- Page limit
- None, and no per page charge
How to use it
- Drop in the scan, the photo or the PDF.
- Choose the language of the document. Getting this right matters more than anything else for accuracy.
- Each page is read in turn, with the progress shown page by page.
- Copy the text out, or download it as a .txt file.
Questions
How accurate is it?
On a flat 300 DPI scan of printed text, high, in the high nineties. On a phone photo taken at an angle in poor light, noticeably worse. On handwriting, poor, because Tesseract is built for print. Photograph the page flat and evenly lit and you will get a usable result.
Does it already know if my PDF has real text?
Yes. It checks for an embedded text layer first and extracts that directly, which is instant and perfectly accurate. OCR only runs on the pages that are genuinely just images.
Why does the language setting matter?
The engine loads a trained model per language and uses that language's letter shapes and word patterns to resolve ambiguity. Reading a French document with the English model turns accented characters into noise.
Is my document uploaded?
No. The language model downloads to you, and the reading happens here. A contract or a medical letter never leaves your device.
Next door
heic to jpg, compress pdf to a size, remove background, passport photo.