Image to Text (OCR)

How to extract text from an image

1. Drop a JPG, PNG, WebP or BMP onto the box above. A preview appears with the image's dimensions and file size, so you can check you picked the right file.
2. Set Text language to the language actually printed in the picture. Twelve are available, from English to Romanian, Russian, Ukrainian, Turkish and Dutch. This choice matters more than anything else you can adjust.
3. Press Extract text. The status line reports each stage — loading the engine, loading the language data, then recognising.
4. Read the result in the box that appears, with a confidence percentage and a character count. Copy it to the clipboard, or download it as a UTF-8 .txt file named after the image.

What people use it for

Retyping is slow and introduces mistakes, especially with numbers. Optical character recognition turns a picture of words back into words you can search, edit and paste — useful any time the text you need exists only as pixels.

• Pulling the figures off a photographed receipt or invoice instead of copying them out by hand.
• Lifting a paragraph out of a screenshot of a chat, a slide or an error dialogue.
• Getting a scanned letter or an old printed page into a document you can actually edit.
• Reading a serial number or a reference code off a photo of a label or a package.
• Extracting a quotation from a photograph of a book page so it can be searched or translated.

Good to know

Two things decide whether the result is useful. The first is the language setting: recognition works from a model of how words in that language are spelled, so pointing an English model at Cyrillic or at Romanian diacritics produces confident-looking nonsense rather than an error. The second is the photograph. Sharp, evenly lit, straight-on text — dark letters on a plain light background — reads very well. Blurred, angled, shadowed or low-contrast text reads badly, and handwriting is not what this engine is built for. The confidence score is your guide: anything under 40% is flagged as unreliable. If a result disappoints, retaking the photo squarely with better light beats trying again with the same file. The output is plain text, so columns and tables are not preserved.

The first time you use a given language, roughly 2 MB of recognition data is fetched into your browser cache; later runs in that language start immediately and work even offline. Your image is a different matter — it is never sent anywhere. That is the whole point, because the things people photograph for OCR are identity cards, signed contracts, payslips and medical letters, and those should not sit in a stranger's upload folder. The picture is read from your disk into this tab, processed by your own device and forgotten when you close it. If your source is a PDF that already contains real text rather than a scan, PDF to Text extracts it exactly and instantly, with no recognition step to get wrong.

Questions people ask

Why is my result complete gibberish?

Almost always the wrong language is selected. Set it to the language printed in the image and run it again. If it is still poor, the picture is too blurry, too dark or too skewed.

Can it read handwriting?

Not reliably. The engine is trained on printed and typed characters. Neat block capitals sometimes come through, cursive generally does not, and the confidence score will show it.

Can I extract text from a PDF here?

Not directly — this tool takes images. For a PDF, try PDF to Text first. If that returns nothing, the PDF is a scan, so convert its pages with PDF to JPG and bring the images back here.

Why does the first run take longer?

The recognition engine and the language data are downloaded and started up once. After that they are cached in your browser and later runs begin straight away.