Account

What new tool do you want?

Image Tools

Image to Text (OCR)

Copy text out of photos, screenshots and scanned images — in English and 100+ languages — with OCR that runs in your browser.

How to use

  1. 1Add, paste or drop an image and pick the language.
  2. 2Press Extract text, then copy or download it.

Frequently asked questions

What images work best?

Sharp, well-lit images with dark text on a light background. Crop to the text area for better results.

Can it read handwriting?

Printed text works best; neat handwriting sometimes works, messy handwriting usually doesn't.

Is my image uploaded?

No — recognition runs on your device. Only the OCR engine is downloaded once.

What's actually running: Tesseract, not a cloud API

Text recognition is powered by Tesseract.js, a WebAssembly build of the open-source Tesseract OCR engine, downloaded and cached in your browser the first time you use a given language. Because it's a real OCR engine rather than a simple pattern-matcher, it handles varied fonts, sizes and some skew reasonably well, and supports around 30 languages from English and Spanish to Arabic, Hindi, Chinese (both Simplified and Traditional), Japanese and Korean — switching languages triggers a fresh download of that language's trained data the first time it's used.

Why the confidence score is worth reading, not ignoring

Every extraction returns a confidence percentage alongside the text, and the tool itself flags anything under 70% with a suggestion to try a sharper or closer crop — that threshold isn't arbitrary, it's the point below which Tesseract's own internal certainty about character recognition starts dropping fast enough that errors (a misread '0' for 'O', dropped punctuation, merged words) become likely. Treating a low-confidence result as a rough draft to proofread, rather than a verbatim transcript, avoids silently propagating OCR mistakes into a document.

Example: pulling a quote out of a screenshot

Say you screenshot a paragraph of text from a PDF or a social media post that doesn't allow text selection, and need the actual words to quote elsewhere. Paste the screenshot directly with Ctrl/Cmd+V (no need to save it as a file first), leave language on English, and press Extract — the recognized text lands in an editable textarea where you can fix the occasional misread character before copying, rather than retyping the whole passage by hand.

Why image quality matters more than engine quality

OCR accuracy depends heavily on contrast and sharpness in the source image, not just the engine behind it — dark text on a light background recognizes far more reliably than text over a busy photo background or a compressed, blurry screenshot. Cropping tightly to just the text region before running OCR, rather than leaving a wide photo with text as a small element inside it, consistently improves results because it removes visual noise the engine would otherwise have to work around.

Staying on English skips a second download

Recognition uses Tesseract's neural-net reader. The first time you press Extract for a given language, that language's data is downloaded, and the status line shows each loading step with a whole-number percentage. A second Extract while Language is still English reuses what was just loaded, so you skip the download. Choosing another language unloads the previous one and fetches the new data — that pause is what you see after switching from English to Arabic or Japanese. During the read itself the button says 'Recognizing text…', and the loader does not print a second copy of that same status. If the download fails, you get 'Couldn't load the OCR engine — check your connection,' and the next press starts the fetch over.

A score of 69.5 is shown as 70, with no warning

The text in the box has blank space removed from the start and the end. Line breaks inside the paragraph stay, so a scanned page keeps its lines. Confidence is a whole percent: 69.5 displays as 70 and the 'try a sharper, closer crop' note stays hidden, while 69.4 displays as 69 and the note is added. Language opens on English, the first of 30 languages. A drop or a paste uses only the first image — two screenshots on the clipboard contribute the first — and clears the previous transcript so old words are not left beside a new picture. The .txt download saves the box after your edits, using the original file name with .txt in place of its extension.

Related tools

More: Image Tools