Image to text converter

How it is calculated

Optical character recognition (OCR) turns the pixels of an image back into letters: a photo of a letter, a scanned invoice or a screenshot of a document becomes text you can copy, search and edit. This image to text converter uses Tesseract, the best-known open-source OCR engine, compiled to WebAssembly and running right in your browser.

How it works

1. Drag in an image, choose one, or paste it with Ctrl + V (for example a screenshot from the clipboard).
2. Check the language in the image. English is preselected; German, French, Spanish and Portuguese are also available.
3. Click “Extract text”. The progress bar shows whether data is still loading or text is already being recognised.
4. Correct the text in the box if needed, then copy it or save it as a .txt file.

Tips for accurate results

• Contrast: dark text on a light, even background works best. Shadows, gradients, patterns or watermarks behind the text reduce accuracy.
• Shoot straight on: photograph the page from directly above, not at an angle. Strongly tilted or curved lines (near a book’s spine) are recognised much less reliably.
• Resolution: around 300 dpi is the usual guideline for scanning. For photos this means the text must be large enough in the frame – capital letters at least about 20 pixels tall. The tool automatically enlarges very small images (up to 2×) and shrinks very large ones to 5,000 pixels on the long side.
• Crop: keep only the area with text – no margins, fingers or table edges. For screenshots, capture the window rather than the whole screen.
• Right language: the language determines the dictionary and character set. With “English”, characters such as ä, ß or é are missing; with “German”, French accents are.

Limitations

Tesseract is built for printed text. Handwriting is hardly recognised, and neither are highly decorative fonts, signs photographed at a steep angle or text on busy backgrounds. Tables and multi-column layouts come out as running text, and the column order may shift. The confidence shown is the average Tesseract reports across all words; below roughly 70 %, a better image is worth it. Always double-check important numbers (amounts, account numbers, invoice numbers) by hand: mix-ups such as 0/O, 1/l or 5/S happen even with good originals.

Data is downloaded the first time

So that recognition works without a server, your browser downloads the recognition engine the first time you click “Extract text” (about 3.7 MB, roughly 1.5 MB compressed in transit) plus the data for the chosen language: English 2.8 MB, German 1.3 MB, French 0.7 MB, Spanish 2.0 MB, Portuguese 1.3 MB. All files are served from this site – no third-party content delivery network is contacted. The browser keeps the language data locally (IndexedDB), so the next run starts much faster. Recognising a page takes a few seconds, depending on your device and the image size.

Privacy

Your image never leaves your device. Recognition runs entirely in your browser (in a web worker), and the extracted text isn’t sent anywhere either. Once you close the tab, nothing is kept – only the downloaded language data stays in the browser’s storage until you clear it.

Frequently asked questions

Sources and legal basis

• Tesseract.js – OCR library for the browser (GitHub, Apache-2.0)
• Tesseract OCR documentation – Improving the quality of the output
• Licences: Tesseract.js, Tesseract engine and language data (Apache-2.0)

As of: 27 September 2026