Image to text — extract text from a photo or screenshot, offline-ready

How to use Image to Text (OCR)

1. Choose one or more images: screenshots, phone photos of a page, or scans.
2. Tick the languages in the image: English, Hindi and/or Telugu. The download size of each is shown beside it.
3. Keep Keep line breaks ticked to follow the layout of the original, or untick it to get one flowing paragraph.
4. Press Extract text. The first run downloads the engine and the language data.
5. Read the result in the Extracted text box, copy it, or save it as extracted-text.txt.

What it does and when to use it

OCR (optical character recognition) turns a picture of words into words you can edit, search and paste. Common uses:

• Copying from a screenshot, such as an error message, a quote from a video, or text in an image you cannot select.
• Typing up a printed page: a notice, a letter, a recipe or a page from a textbook.
• Pulling details off a receipt or bill to paste into a spreadsheet.
• Reading a Hindi or Telugu notice into text you can translate or search.

Because nothing leaves your device, it is also suitable for documents you would rather not upload, like letters with your address on them.

How it works

The tool uses Tesseract, a long-standing open-source OCR engine, through tesseract.js (Apache-2.0).

1. Downloads. On first use the browser fetches the Tesseract worker and core (about 3.9 MB) and the trained data for each language you ticked, from jsDelivr. The language files are stored in your browser’s IndexedDB, so they are not fetched again next time.
2. Recognition. Tesseract’s neural-network recogniser finds lines of text, then reads each word. With more than one language ticked, it considers all of them together.
3. Several images. Each file is read in turn. With more than one image, each block of text starts with the file name.
4. Result. You get the text, the number of characters, the languages used and an average confidence score.

The image itself never leaves the page. The engine is shut down when the job ends to free memory.

Worked examples

A screenshot of an error dialog. Choose the screenshot, keep English, and press Extract text. Screenshots are sharp and evenly lit, so they are the easiest case. The text appears with its line breaks, ready to paste into a search engine or a support ticket.

A photo of a two-column printed notice. Tesseract reads columns as separate blocks, but the order can come out mixed. If that happens, crop each column with the Image Cropper and run them one at a time.

A Hindi circular photographed on a phone. Untick English, tick Hindi, and run it. If the confidence is low, take the photo again straight on, in daylight, with the page filling the frame. A tilted page can be straightened first with the Image Rotator.

Limits and tips

• Sharp and straight wins. Hold the phone parallel to the page, fill the frame, and avoid shadows from your hand.
• Dark text on a light background reads best. White text on a dark or busy background often fails; try the Image Invert tool first.
• Small text needs pixels. If the letters are only a few pixels tall, take a closer photo rather than zooming a screenshot.
• Tables lose their shape. The text comes out, but columns may not line up. Check numbers carefully before you rely on them.
• Always proofread. Even at high confidence, OCR can swap similar shapes like 0 and O or 1 and l. Do not use the output for anything important without checking it against the image.
• Three languages only. Other scripts are not supported on this page yet.
• Going the other way? To write corrected text onto a picture, use the Image Text Overlay tool.