What it reads well
Printed text: university and NTA notices, book pages, question papers, typed letters and screenshots, in Hindi (Devanagari), English and mixed Hindi-English, and in Sanskrit, Marathi, Nepali, Bengali, Assamese, Gujarati, Punjabi, Odia, Tamil, Telugu, Kannada, Malayalam and Urdu. Handwriting and fancy display fonts are read poorly by any OCR.
For a PDF that already has text in it (most PDFs made on a computer), the text is copied straight out of the PDF, which is exact. The tool checks that text first: pages typed in Kruti Dev, DevLys or another old Hindi font (whose text copies as Latin letters, like "vkSj" for "और") and pages with broken characters are read as images instead, so you get Unicode Hindi. Tick the option to read every page as an image if the copied text still comes out wrong.
For the best result
- Photograph the page straight on, in daylight, with the whole page in the frame and no shadow of your phone.
- Crop away the table or the floor around the page.
- Pick the right language: Hindi and English together is slower than one language but reads mixed notices best.
- Keep "Even out shadows" on for phone photos: a shadowed or dim page is turned white before it is read, and small text is enlarged. Screenshots and clean scans are read as they are.
- Press "Join lines" after reading to turn the broken lines of each paragraph back into one line, ready for Word or WhatsApp.
Hindi text you can edit
The Hindi comes out as Unicode Devanagari: paste it into Word, Google Docs, WhatsApp or a Mangal-font document and edit it. Check the matras and the half letters (क्ष, त्र, ज्ञ, ड़) and every number against the original: OCR can mistake 0 for o, 1 for l, or drop a dot.
Private by design
The reading runs inside your browser with Tesseract.js, the open-source OCR engine; your images are never uploaded. The engine and the language files (a few MB) come once from the jsDelivr network, and your browser keeps them for the next time.