Image to Text (OCR)
Extract text from screenshots, photos and scanned pages — recognised privately in your browser. English, German, French and Spanish text.
Preview
Settings
The first run downloads the OCR engine and English model (about 9 MB) from this site; your browser then keeps them for next time.
Add an image to get started. The recognised text appears here.
Result
Check these words (under 70% confidence — click to find)
Related Tools
How to Use
- Add an image — Choose image files, drop them on the page, or paste a screenshot with Ctrl+V (⌘V on a Mac). PNG, JPG, WebP, GIF and BMP work in every browser; you can add several images at once.
- Check the preview — Rotate the image if it is sideways. For faint, coloured or low-contrast pictures, try grayscale, more contrast or a black-and-white threshold, and use "Invert" for light text on a dark background. "Enlarge small images" helps with small screenshots.
- Extract the text — Choose the language of the text (English, German, French or Spanish), then click "Extract text" (or "Extract all" for several images). The first run downloads the OCR engine and the model for that language once; after that, recognition starts immediately.
- Review, copy or save — The result shows an overall confidence score and lists words the engine was unsure about — click one to find it in the text. Edit as needed, join broken lines into paragraphs, then copy or download a .txt file.
About Image to Text (OCR)
Image to Text turns pictures of text — screenshots, phone photos of documents, scanned pages, slides, signs and receipts — into editable plain text using optical character recognition (OCR). It runs entirely in your browser: the images are processed on your device and never uploaded.
How it works
- Decoding. Your browser decodes the image, applying the rotation stored by phone cameras (EXIF orientation), so photos appear the right way up.
- Resizing. Tesseract reads best when capital letters are roughly 20–40 pixels tall. Images whose long side is under 1,000 pixels are enlarged (up to 3×) and very large images are reduced to at most 4,000 pixels and 12 megapixels.
- Optional clean-up on a canvas: grayscale; contrast; invert for light-on-dark text; and threshold, which turns every pixel black or white. "Auto" picks the threshold with Otsu's method, which finds the brightness level that best separates text from background; "Manual" lets you set it. Transparent areas are flattened onto white.
- Recognition. The prepared image goes to Tesseract (via tesseract.js 7) running in a background Web Worker as WebAssembly, using its LSTM neural-network recogniser and the model for the language you chose — English, German, French or Spanish — from Tesseract's "best" training set (integer version). The layout setting tells Tesseract whether to expect a full page, a single column, one block, one line or scattered text.
- Results. You get the text, Tesseract's average confidence, and a list of low-confidence words. The text keeps the line breaks from the image; "Join broken lines" merges the lines inside each paragraph.
Choosing a layout
| Layout | Use it for |
|---|---|
| Automatic | Documents and pages with headings, paragraphs or several blocks — the default. |
| Single column | Articles, book pages and long screenshots of one column of text. |
| Single block | One paragraph or a tightly cropped area of text. |
| Single line | A title, a code, a serial number or one line of a receipt. |
| Sparse text | Screenshots of apps, forms, diagrams, signs and labels where text is scattered. |
Tips for photos and screenshots
- Photograph pages straight on, in good even light, and fill the frame with the text.
- Crop to the part you need before adding the image — less clutter means fewer stray characters.
- Screenshots are usually perfect as they are; leave clean-up off and keep "Enlarge small images" on.
- For dark mode screenshots, white text on black, or chalkboards, turn on Invert.
- For yellowed paper or shadows across the page, try Threshold: Auto, then adjust it manually if letters break up or fill in.
Limits
OCR is never perfect. Expect mistakes with blurry, tilted or very small text, decorative fonts, handwriting, text on busy backgrounds, and similar-looking characters such as 0 and O or 1, l and I. Tables and multi-column layouts come out as lines of text without the original structure. Four languages are supported — English, German, French and Spanish — and the selected language matters: with the wrong one, letters such as ä, ß, é, ç and ñ are often misread. For a page that mixes languages, choose the main one. Check important text — names, numbers and amounts — against the original.
Who uses it
Students copying quotations from textbook photos and lecture slides, office workers turning printed letters and receipts into editable text, researchers digitising archive pages, developers grabbing error messages from screenshots, and anyone who needs text that is trapped in a picture. To extract text from a PDF that already contains a text layer, the PDF to Text Converter is faster and exact; to tidy the result, try Remove Line Breaks.
Privacy and offline use
The OCR engine (about 4 MB) and the language models (English about 5 MB, German about 2 MB, French about 1.2 MB, Spanish about 3.4 MB) are hosted on this site and downloaded only when you first use them; your browser then keeps them in its cache. Images, text and results never leave your device and are discarded when you close the page.
Frequently Asked Questions
Are my images uploaded?
No. Text recognition runs in your browser with the open-source Tesseract engine compiled to WebAssembly. Your images and the extracted text stay on your device, so the tool is safe for receipts, letters, IDs or confidential screenshots.
Which languages are supported?
English, German, French and Spanish. Choose the language of the text under "Text language" before you extract: each language has its own Tesseract model, which also reads digits and common punctuation, and German umlauts and ß, French accents and ç, and Spanish ñ, accents and ¿ ¡ come out right only when the matching language is selected. Text in other Latin-script languages is often partly readable, but accented letters may be missed, and other scripts such as Chinese, Arabic or Cyrillic are not recognised.
Why is the first run slower?
The first time you extract text, your browser downloads the OCR engine (about 4 MB) and the model for the selected language (English about 5 MB, German about 2 MB, French about 1.2 MB, Spanish about 3.4 MB) from this site. They are then kept in your browser's cache, so later runs and repeat visits start much faster. Starting the engine still takes a few seconds on slower devices.
How can I get more accurate results?
Use the sharpest, largest image you have: text should be at least 20 pixels tall, straight and evenly lit. Crop away unrelated parts, rotate sideways images, and try the clean-up options — grayscale and extra contrast for faded or coloured text, a threshold for uneven backgrounds, and Invert for light text on dark backgrounds. For a single line or a label, choose that layout.
What does the confidence score mean?
It is Tesseract's average certainty across all recognised words, from 0 to 100. Above about 85 the text is usually close to perfect; between 65 and 85 expect some mistakes; below that, check the image quality. Words under 70 percent are listed separately so you can check them first.
Can it read handwriting?
Only neat, printed handwriting, and not reliably. Tesseract is trained on printed text, so cursive and quick notes usually come out garbled. Typed or printed documents, screenshots, signs and book pages work best.
Can I extract text from a scanned PDF?
Not directly. First try PDF to Text — if the PDF has a text layer it is faster and exact. For a scanned PDF, save or screenshot each page as an image (most PDF viewers can export pages as PNG or JPG) and add the images here.
Is there a size limit?
Images up to 50 MB can be added. Very large photos are reduced to at most 4,000 pixels on the long side (and 12 megapixels) before recognition, which keeps memory use reasonable without losing readable detail; small images are enlarged up to three times because Tesseract reads small text poorly.