Skip to content
CornerTools

Image to Text

Copy the text out of a photo or screenshot — English, Hindi and other Indian languages — without uploading it.

Drop images here or paste them, or

Your images never leave your device

Processed in this browser · JPG, PNG, WebP, GIF, BMP, AVIF or HEIC · up to 30 at a time

Language

The text reader downloads the first time it is needed, 11.97 MB for English; your browser keeps it for later visits.

Words the reader was unsure of are tinted; check them before you use the text.

Copies the text out of screenshots, photos of documents and scanned pages, so you can paste, search or edit it instead of retyping it. It reads 15 languages, Hindi and nine other Indian languages among them, marks the words it was unsure of so you know what to check, and runs on your device: the images are never uploaded.

How to use it

Paste a screenshot (Ctrl+V, or ⌘V on a Mac), drop images on the page, or choose them; up to 30 at a time, iPhone HEIC photos included. Each image is read in turn, its text shown beside it, ready to edit, copy or download as a .txt file. With several images, Download all saves every image's text in one file, each under its file name.

  • Language: English, Hindi, Marathi, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi, Urdu, Spanish, French, German and Portuguese. Add up to 3 for an image that mixes them, such as a Hindi form with English words. When the text is in a script none of your languages is written in, the tool offers to read it in the language it looks like, as “Read as Hindi” for Devanagari.
  • Select area: reads only the part of an image you mark, which skips menus and photos around the text. The same box turns and straightens a photo taken at an angle.

The text reader downloads from this site the first time you add an image: about 11.97 MB for English, and a little more for each language you add. Your browser keeps it, so later visits start reading at once.

How it works

The text is read by Tesseract, an open-source OCR engine whose neural network reads a whole line at a time. It is trained on dark print on light paper at book sizes, so each image is prepared to look like that first. A mostly dark image, such as a dark-mode screenshot, is inverted. Small text is scaled up: the height of the image's lines of text is measured, and lines under 24 pixels tall are enlarged towards 40 pixels, up to 4×, the size at which the engine reads best. The image is then turned upright if it is sideways or upside down, and read as a 300 DPI page.

The engine scores each word from 0 to 100 for how sure it is. Words scoring under 70 are tinted in the result: they are where misreadings collect, such as a 0 read as O or a missed accent. Once you edit the text the tint goes, as the words have moved.

Frequently asked questions

Not reliably. The engine is trained on printed and on-screen text; neat block capitals sometimes read, joined-up writing rarely does.
Give it the sharpest, straightest image you have: a screenshot rather than a photo of a screen, and a photo taken square to the page in good light. Choose only the languages the text is in, since each extra one gives the engine more ways to go wrong. Mark the area with the text, so pictures and borders are not read as letters.
Each row comes out as a line of text, its cells parted by spaces, so the columns no longer line up. Grid lines drawn around the cells are left out before the table is read, as they would mislead the reader.
PDF to Text copies the text out of a PDF, reading scanned pages the same way, and Searchable PDF (OCR) makes a scanned PDF searchable.
No. Your browser opens the images and reads their text on your device; they are not sent to CornerTools or anyone else. To check, open the browser's developer tools at the Network tab and add an image: no request carries it. The text reader and its language models download from this site the first time: a download to your browser, not an upload of your image.

See also