Image to Text (OCR) — Extract Text from Any Image

Guide: getting accurate text out of photos and screenshots

How recognition runs

The tool uses tesseract.js, an open-source OCR engine compiled to run in the browser. It loads when you click Extract text. The first time you use a language, the engine and that language's recognition data are downloaded from a CDN and cached, so the first run is noticeably slower than later ones. Your image itself is read on your device and is not uploaded.

The recognised text appears in an editable box. Copy text copies whatever is in the box, including your corrections. Download .txt saves the text as it was first recognised, so if you fix mistakes in the box, use Copy text to keep them. The tool accepts any image your browser can open, such as JPG, PNG or WEBP.

Choosing the language setting

There are three options: English, Hindi, and English + Hindi. Use English for English-only text and Hindi for text in Devanagari. For mixed documents, such as Hindi forms with names written in English, bilingual government notices or shop bills, choose English + Hindi; it loads both sets of data and takes a little longer. Other scripts, such as Gujarati, Bengali, Tamil or Urdu, are not supported, and the output for them will not be usable.

Preparing the image for better accuracy

• Crop to the text you need with the Crop tool. Logos, photos and borders around the text add noise.
• Straighten tilted pages with the Rotate tool. Lines that slope confuse the engine.
• Photograph documents in even light, without your phone's shadow across the page.
• Hold the phone parallel to the page and make sure the text is in focus.
• For newspapers and multi-column layouts, crop one column at a time so lines are not mixed together.

Everyday uses

• Copying text from a screenshot of a circular, notice or exam schedule.
• Typing up printed notes or textbook passages for an assignment.
• Pulling amounts and dates from bills into an expense sheet.
• Reusing a quote or address from a poster or visiting card.
• Reading a scanned PDF: convert the page with PDF to Image first, then extract the text here.

What OCR commonly gets wrong

Handwriting is recognised poorly, and decorative fonts are often misread. Tables come out as plain lines of text, so column alignment is lost. Similar characters are easily swapped: 0 and O, 1, l and I, and 5 and S. In Hindi, matras and conjunct letters are sometimes misread, especially in small or low-quality print. The rupee symbol and other special characters may be dropped or replaced.

Very small text is another common cause of errors, for example a phone screenshot of a long document. Zooming in on the original and taking a fresh screenshot, so the letters appear larger, often helps more than changing any setting.

Before you rely on the text

• Check every number, such as account numbers, amounts, dates and roll numbers, against the image.
• Read names and addresses carefully; the engine does not know which spelling is correct.
• If you corrected the text in the box, copy it rather than downloading the .txt file.
• If the result is mostly wrong, retake the photo instead of correcting it line by line.

Related guides

• Best Free Online Image Tools for Students
• How Professional Designers Edit Images Faster