Skip to content

Image to Text (OCR)

Turn a photo or screenshot of a page into editable text on your device. The recognition engine loads only when you click Extract text, and nothing is uploaded.

Or drop a photo here, or paste with Ctrl+V / ⌘V. Nothing is uploaded.

English is ready after the engine loads. Other languages download a language pack from this site the first time you use them, not from a third-party CDN.

Image

No image

Recognition is never perfect. Proofread numbers, names and punctuation before you rely on the result.

What this tool is for — and what it isn't

Use it when you have a picture of printed text: a signed letter, a slide, a book page, a screenshot of an error, a scanned form. It will try to type that text so you can copy it or save a .txt file. It is not a human typist and it is not a document-layout app. Expect to fix mistakes, especially in numbers and proper names.

How to extract text from an image

  1. Add an image. Use Select, drop a file onto the box, or paste a screenshot with Ctrl+V / ⌘V. JPG, PNG, WebP, GIF and BMP are accepted. HEIC photos from an iPhone should be converted to JPG first.
  2. Choose the language. Pick the language the page is written in. English is the default. Spanish, French, German, Portuguese, Arabic, Urdu and Hindi load from this site the first time you use them.
  3. Extract and check. Click Extract text. The first run on a device loads the recognition engine (a few megabytes) and then the language data; later runs are faster. Proofread the result, then copy it or download a .txt file.

Accuracy, honestly

Modern OCR is good on clean, high-contrast printed type and much less reliable on everything else. A well-lit, straight-on photo of a printed page in the language you selected can come out almost complete. Handwriting, decorative fonts, low-resolution phone snaps, skewed pages, stamps and watermarks will lose words or invent them. The confidence figure is the engine's own estimate for the whole page — useful as a hint, not as a score you can quote.

Tips that actually help

  • Crop to the text so the engine isn't spending time on a desk, a margin or a thumb.
  • Use even lighting and avoid glare. A shadow across a line is a common source of errors.
  • Straighten the page, or take the photo from directly above. A strong angle distorts letter shapes.
  • Give it enough pixels. A few hundred pixels of height for a paragraph is tight; a larger photo (around 300 DPI if you are scanning) is better than a tiny thumbnail.
  • Dark text on a light background works better than light-on-dark. Invert or retake if you can.
  • For multi-column layouts, crop and run one column at a time.
  • Pick the correct language before you run. Arabic and Urdu are written right to left; the result is Unicode text you can paste into a document.
  • To count the words you extracted, paste the text into the word counter.

How it runs in your browser

When you click Extract text, the page lazy-loads Tesseract.js, then a WebAssembly build of the Tesseract engine and the language pack — all from this site under /ocr/. Other pages on the site do not load any of that. The first run is the slow one; after the files are cached, later images in the same language are usually quicker.

Related tools and guides

Image to text FAQ

Are my images uploaded?

No. The picture is read in your browser. The recognition engine and language files are loaded from this site, not sent to a third-party server with your image. Nothing is stored on our servers.

How accurate is this?

It is useful on clear, printed or typewritten text that is reasonably sharp, high-contrast and roughly upright. It is much weaker on handwriting, stylised display fonts, very small type, strong perspective, heavy shadows, or a photo taken at an angle. The confidence percentage is a rough engine score, not a guarantee. Always proofread, especially numbers, names and punctuation.

Which languages can I use?

English, Spanish, French, German, Portuguese, Arabic, Urdu and Hindi. The engine and English data load when you first click Extract text. Other languages download their own pack from this site the first time you pick them. Mixed-language pages work less well than a page in a single language; pick the main language.

Does the first run use a CDN?

No. The worker, the WebAssembly engine and the language packs are hosted on this site under /ocr/. Your browser may cache them after the first use so later visits are quicker.

What image types can I open?

JPG, PNG, WebP, GIF and BMP, and AVIF if your browser can display it. Animated GIFs are read as a single frame. iPhone HEIC photos usually need converting first — use HEIC to JPG, then come back here.

Can I extract text from a PDF?

Not directly on this page. If the PDF is a scan (a photograph of a page, not selectable text), convert the pages to images with PDF to JPG and run those images through this tool, one page at a time. A PDF that already contains a text layer should be copied in a reader instead — OCR would only degrade it.

Will it read handwriting?

Occasionally, if the writing is very neat and high-contrast, but you should not rely on it. This engine is trained mainly on printed type. Treat any handwritten result as a draft.

What about tables, columns or receipts?

Layout is flattened into reading order, which is usually left-to-right, top-to-bottom (or the equivalent for Arabic and Urdu). Multi-column pages, dense tables and till receipts often come out with columns mixed together. Crop to one column or one block of text and run it separately for a cleaner result.

Why is the first click slow?

The engine (several megabytes of WebAssembly) and the language pack have to download from this site and start up. Progress is shown while that happens. After that, the same language should run more quickly on this device.

Does it work in every browser?

It needs a current browser with WebAssembly and Web Workers: recent Chrome, Edge, Firefox or Safari. Very old browsers, or a strict private mode that blocks workers, will fail to load the engine.

What can I do to get a better result?

Crop to the text, use even lighting, straighten the page, and aim for a reasonably large photo (around 300 DPI is a good rule of thumb). Dark text on a light background works better than the reverse. If the text is in another language, select that language before you run it.