Extract printed text from an image with OCR in English, Urdu, Arabic, French, German or Spanish, then edit, copy or download it as .txt. Runs in your browser.
If this tool isn’t working as expected, please take a screenshot of the error and report the problem here so we can investigate and improve it.
OCR solves a specific problem: text that's visually present in an image but not selectable as real characters. A screenshot of an article, a photo of a printed page, a scanned receipt, or a slide exported as an image all lock their text inside pixels - you can't select, search, or copy it directly.
This tool runs Tesseract.js, an open-source OCR engine, against your image entirely in the browser, so the image itself never leaves your device - a real advantage for receipts, forms, or other sensitive material. After recognition the text appears in an editable box, so you can correct mistakes before using it, with a recognition confidence figure and a count of words the engine was unsure about to show you where to look.
Copy it, or download it as a UTF-8 .txt file named after your image.
This OCR (optical character recognition) tool reads printed or typed text from an image - screenshots, scanned documents, receipts, or photos of pages - and turns it into text you can edit, copy or save as a .txt file. Choose the language of the text first: English, Urdu, Arabic, French, German or Spanish. Recognition runs on Tesseract.js directly in your browser, using engine and language files served by this site itself, so your image and the extracted text are never uploaded anywhere. Like all Tesseract-based OCR, it is built for machine-printed and typed text - handwriting recognition is unreliable and not a supported use case.
Upload an image containing printed text such as documents, screenshots, or photos.
The OCR engine scans the image and extracts readable text automatically.
Review the extracted text for accuracy before copying it - the result is read-only in this tool.
Copy the extracted text to clipboard and use it anywhere instantly.
Choose a JPG, PNG, WebP or other image, or drag one onto the page. Use Replace image to switch to another.
Pick the language the text is written in. Changing it later doesn't start a new run by itself - click Extract again.
Click Extract Text. Steps and a progress bar show the engine loading, the language loading and the reading.
Review the confidence figure, correct the text in the box, then Copy Text or Download .txt.
Best For
Input
receipt.jpg - a photo of a printed receipt, Language set to FrenchOutput
Editable text in the result box, a recognition confidence figure, and a download named receipt-ocr.txtRecognition runs locally with the French language data. Fix any misread words in the box before copying or downloading; the downloaded file matches what is in the box.
Choose the receipt's language, extract, check the low-confidence words, and copy or download the text.
Extract text from error messages, article screenshots or slides exported as images.
Select Urdu or Arabic so the right language data is used; the result is shown right-to-left and downloads as UTF-8 text.
Problem
Solution
Text in another language is misread badly with the wrong language selected. Choose the language of the text, then click Extract Text or Extract again.
Problem
Solution
Tesseract.js is built for machine-printed and typed text. Handwriting recognition is unreliable and will often produce garbled or incomplete results.
Problem
Solution
Recognition confidence shows how sure the engine was about the letters it chose. A high figure doesn't guarantee correct text, so read the result before using it.
Problem
Solution
Only one language is used per run. For a page with two languages, extract once for each and combine the text.
Sharp, high-contrast printed text gives the best results. Blurry, skewed or low-resolution images give lower confidence.
The result shows how many words the engine was unsure about. Numbers, names and unusual spellings are the usual ones.
The text box is editable, and Copy Text and Download .txt both use exactly what is in it.
A photo saved sideways with a rotation tag is turned upright before it is read, and very large photos are scaled down first so the browser stays responsive.
English, Urdu, Arabic, French, German and Spanish are supported. Each run reads one language; there is no automatic language detection and no translation.
Tesseract.js, the OCR engine used here, is designed for machine-printed and typed text. Handwriting recognition is unreliable and not a supported use case.
The result is plain text. Columns, tables and formatting are not preserved, and there is no PDF or Word export - only Copy Text and a .txt download.
To keep recognition fast and your browser responsive, an image larger than about 4000 pixels on its longest side or about 12 megapixels is scaled down before it is read, and the tool tells you when it did. An image of more than about 150 megapixels is refused. Your original file is never changed.
The engine (about 3 MB) and the chosen language (about 1 to 3 MB) are served by this site, not a third-party CDN, and your browser keeps them for later runs. You need to be online for that first download. If a download fails or stalls, the tool says which language could not be loaded and a run that takes longer than two minutes is stopped; click Try again.
Tesseract.js is built for one of these two inputs.
| Printed or typed text | Handwriting | |
|---|---|---|
| Recognition engine fit | Tesseract.js is trained and built for this - the intended use case | Not a supported use case - Tesseract-based engines aren't designed for handwriting |
| Typical accuracy | Generally reliable on clear, well-lit source images | Unreliable - expect missing words, garbled characters, or empty results |
Which should you use?
Use OCR image to text for printed and typed text. For handwriting, expect to retype most of it.
The most common questions are which languages are supported and whether the image is uploaded. Six languages are supported, the image is never uploaded, and the OCR files come from this site rather than a third-party CDN. The FAQ below also covers confidence, editing, saving as .txt and large photos.
No. Text recognition runs in your browser using Tesseract.js, and your image and the extracted text are never sent anywhere. The OCR engine and the language data (about 1 to 3 MB per language) are served by this site itself rather than a third-party CDN, and your browser caches them after the first run.
English, Urdu, Arabic, French, German and Spanish. Choose the language of the text before clicking Extract Text; each run reads one language. There is no automatic language detection or translation.
Recognition confidence shows how sure the engine was about the characters it chose, from 1 to 100%, labelled High, Medium or Low. It is not the same as accuracy, so always read the result. Where the engine is unsure about individual words, the tool also tells you how many, so you know where to check. If the engine returns no usable confidence, the tool says Not available instead of showing 0%.
Yes. The result appears in an editable box, so you can fix recognition errors. Editing never re-runs OCR, and Restore OCR text brings back the original result. If you run Extract again, the text in the box is replaced.
Yes. Download .txt saves exactly the text in the box (including your edits) as a UTF-8 text file named after your image, for example receipt.jpg becomes receipt-ocr.txt. Urdu, Arabic and accented letters are preserved. Copy Text puts the same text on your clipboard.
Not reliably. Tesseract.js, the OCR engine behind this tool, is built for machine-printed and typed text. Handwriting recognition is a different problem this tool doesn't target, so results on handwritten notes are often garbled or incomplete.
A photo larger than about 4000 pixels on its longest side or about 12 megapixels is scaled down before it is read, and the tool tells you so. Your original file is not changed. Photos saved sideways with a rotation tag are turned upright first. An image of more than about 150 megapixels is refused with a message.
The upload step rejects anything that isn't an image file. If a valid image fails to produce text, common causes are the wrong language selected, low resolution, heavy blur or glare, handwritten content, or a genuinely blank image - try the right language and a clearer printed-text image. If the page says a language (for example Urdu) could not be loaded, click Try again. A run that takes longer than two minutes is stopped with a message, and your image stays in place.
Yes. Drag one image onto the upload area or onto the page, or use Replace image at any time. Replacing the image clears the old text and confidence. Clear removes the image and the text.
Leave your email so we can prioritize similar tools and updates.