Image to Text (OCR)

An image to text converter, often called an OCR tool, reads the words in a picture and turns them into text you can copy, edit and search. Our free tool runs optical character recognition entirely in your browser: drop in screenshots, photos of documents or scanned pages, or paste an image straight from your clipboard, and the text appears in a few seconds. Nothing is uploaded, so invoices, contracts, ID documents and client emails stay on your device. It currently recognises English text only, shows a confidence score so you know how much checking to do, and lets you copy the result or download it as a plain .txt file.

How to use the Image to text

1. 1 Add your imagesDrop up to ten images onto the tool, choose them from your device, or press Ctrl+V (Cmd+V on a Mac) to paste a screenshot you have just taken. PNG, JPG, WebP, GIF and BMP all work.
2. 2 Extract the textPress Extract text. The first time, your browser downloads the OCR engine and the English language data from this site, which takes a few seconds; after that it is cached.
3. 3 Watch the progressThe progress bar shows each stage, from loading the engine to reading the text. Large, detailed images take longer than small screenshots.
4. 4 Check the confidence scoreEach result shows a word count and a confidence score. A low score means the image was hard to read, so proofread carefully, especially names and numbers.
5. 5 Copy or downloadCopy the text to your clipboard, download it as a .txt file, or, if you processed several images, download all of the text in a single file.

What OCR is and how this tool reads your images

Optical character recognition (OCR) is the process of finding letters and words in a picture and turning them into real, editable text. To a computer, a screenshot or a photo of a letter is just a grid of coloured dots; OCR software looks for the shapes of characters, groups them into words and lines, and outputs text you can paste into an email, a spreadsheet or a document.

This tool uses Tesseract, a long-established open-source OCR engine originally developed at Hewlett-Packard and later maintained with support from Google. Modern versions of Tesseract use a neural network trained to recognise whole lines of text rather than single letters, which makes them far better at ordinary printed text than older OCR. We run a WebAssembly build of Tesseract, called Tesseract.js, inside your browser. The engine and the English language model are served from our own website, so the tool does not contact any other service, and the image you choose is processed in your browser's memory and never sent anywhere.

Before recognition starts, small images are scaled up and every image is placed on a white background. Tesseract works best when letters are at least 20 or so pixels tall, so enlarging a small screenshot often improves accuracy noticeably.

Everyday uses for small businesses and freelancers

Retyping text from an image is slow and error-prone. An image to text converter saves time whenever words are trapped in a picture, for example:

• Copying the details from a photographed receipt or supplier invoice into your bookkeeping software.
• Pulling an error message out of a screenshot so you can search for it or paste it into a support ticket.
• Turning a photo of a printed price list, menu or flyer into text you can edit and reuse.
• Extracting quotes or paragraphs from a scanned page of a book or report for research notes.
• Getting the text out of a slide or a webinar screenshot when the original file is not available.
• Making the words in an image searchable by saving them alongside the picture in your files.

How to get the most accurate results

OCR accuracy depends far more on the image than on the software. Clean, sharp, high-contrast images of printed text usually come out almost perfectly; blurry phone photos taken at an angle in poor light can produce a lot of mistakes. A few habits make a big difference.

Photograph documents straight on, with the page flat and filling as much of the frame as possible, in even light without shadows or glare. Crop the image to the text you need, because logos, photos and decorative borders can confuse the engine and create stray characters. Screenshots are ideal sources because the text is perfectly sharp, but very small interface text benefits from zooming in before you take the screenshot.

Dark text on a light background works best. White text on a dark or coloured background, text over photos, and very light grey text are harder. If a result looks poor, try increasing the contrast or converting the image to black and white in a photo editor, then run it again.

Layouts matter too. Simple paragraphs come out in reading order. Multi-column pages, tables and forms are read line by line, so columns may be merged and table cells may lose their structure. For tables, it is often quicker to extract the text and then tidy it in a spreadsheet than to retype everything.

Understanding the confidence score

After reading an image, Tesseract reports a mean confidence score between 0 and 100. It is the engine's own estimate of how sure it is about the words it recognised, averaged across the whole image. We label scores of 85 and above as high, 65 to 84 as fair and anything lower as low.

A high score is a good sign but not a guarantee. The engine can be confident and still wrong, for example reading the letter O as the number 0 in a reference code, or the letter l as the number 1. A low score is a clear warning that the image was difficult and that you should compare the text with the original carefully. Whatever the score, always check names, figures, dates, email addresses, bank details and anything you will rely on legally or financially.

Language support and what the tool cannot read

The tool currently recognises English only. It will usually read other languages that use the same Latin alphabet, such as French or German, but accented letters may be dropped or replaced, and spelling-based corrections are tuned for English. Text in other scripts, such as Arabic, Chinese, Urdu or Cyrillic, will not be recognised correctly.

Tesseract is designed for printed and typed text. Handwriting, especially joined-up handwriting, is generally read poorly or not at all. Very stylised fonts, text on curved surfaces, number plates, CAPTCHAs and text in low-resolution video frames are also unreliable. If you need to work from PDF files rather than images, use our PDF to text tool first, as many PDFs already contain a text layer that can be extracted exactly without OCR.

Privacy, speed and offline use

Many online OCR services ask you to upload your image to their servers. That may be acceptable for a meme, but not for a passport, a payslip, a medical letter or a customer's details. Because this tool runs in your browser, the image is never uploaded and we never see it. That makes it a sensible choice when you are handling personal data and want to keep to the data minimisation principle in UK GDPR.

The first time you use the tool, your browser downloads around 6 MB of files from this site: the OCR worker, the WebAssembly engine and the English language model. The language model is stored in your browser's own storage, so later visits start much faster, and once the page and the engine have loaded the recognition itself works without an internet connection.

Processing time depends on your device and the size of the image. A typical screenshot is read in a few seconds on a modern laptop; a full-page, high-resolution photo can take longer, particularly on older phones. Very large images are scaled down to keep memory use reasonable. If you have many pages to process, run them in batches of up to ten.

Examples you can copy

Works well                         Works poorly
---------------------------------  ---------------------------------
Screenshots of web pages and apps  Handwritten notes
Flat, well-lit photo of a letter   Photo taken at a steep angle
Scanned printed pages (300 dpi)    Blurry or low-resolution images
Dark text on a white background    Text over photos or patterns
Simple paragraphs                  Complex tables and multi-column layouts

O (letter)  <->  0 (zero)
l (letter)  <->  1 (one)  <->  I (capital i)
S           <->  5
B           <->  8
rn          <->  m
"           <->  ''   (straight and curly quotes)

Common mistakes to avoid

Frequently asked questions