Image to text

Extract text from an image

Read the words out of a photo, a screenshot or a scan — and get text you can select, correct and paste.

Extract text from an image online

A photograph of a page, a screenshot of a message, a scan of a receipt: the words are there but you cannot select them, because the file is a picture. Optical character recognition looks at the shapes and works out which letters they are, giving you back editable text. This tool runs that recognition with Tesseract compiled to WebAssembly, entirely inside your browser — the image is opened locally and never uploaded, which matters for a payslip, an identity document or a medical result. Two language settings are offered, English with French as a secondary, and French with English as a secondary; the corresponding model, five to fifteen megabytes, downloads on first use and is then reused. Accuracy depends on the picture: sharp, well-lit, straight-on text at a reasonable resolution reads well, while a blurred photo taken at an angle in poor light will produce mistakes you have to correct. When little or nothing is found, the tool says so rather than returning silence.

How to extract text from an image

  1. Choose the image

    A photo, a screenshot or a scan. It stays on your device — the recognition runs in your browser, not on our servers.

  2. Pick the language

    English or French as the primary, the other as secondary. The first run downloads the model, five to fifteen megabytes, and later runs reuse it.

  3. Read and correct the text

    The recognised text appears ready to select and copy. Check the passages OCR finds hard: unusual names, figures, and anything printed small.

Why use it

The image never leaves your device

Recognition runs in the browser with WebAssembly, so a payslip, an identity document or a medical result is not uploaded anywhere — including to us.

Retype nothing

A reference, an address or a paragraph photographed on a screen becomes text you can paste, instead of something you copy out by hand and get wrong.

Says when it found nothing

An image with no readable text produces a clear warning suggesting another language or a sharper picture, rather than an empty box you have to interpret.

Frequently asked questions

Is my image uploaded?
No. Recognition runs entirely in your browser through Tesseract compiled to WebAssembly. The image is read locally and never sent to our servers, which is why the tool is suitable for a document you would rather not upload anywhere.
Which languages are supported?
Two settings: English with French as a secondary, and French with English as a secondary. Pick the one matching the dominant language of the image — choosing the wrong one noticeably degrades the result, particularly on accented text.
Why is the first run slower?
Because the language model — five to fifteen megabytes — is downloaded to your browser on first use. It is kept afterwards, so later extractions start immediately.
How accurate is it?
It depends on the image, and the difference is large. Sharp, well-lit, straight-on printed text at a reasonable resolution reads well. A blurred photo, a slanted angle, poor light, a busy background or handwriting will produce errors. Always reread the figures and the proper nouns — those are where OCR fails most often and where a mistake costs most.
Can I use it on a PDF?
This tool takes images. For a scanned PDF, use the OCR tool, which handles the document page by page and gives it a text layer.
Do I need an account?
Yes. The recognition itself is local, but the tool sits inside the signed-in workspace. Creating an account is free.

Related tools