Image to Text (OCR)

No. Recognition runs in your browser using Tesseract, the open-source OCR engine, compiled to WebAssembly. The engine and the language files are served from this site, not a third-party CDN, and the images themselves are never transmitted. That makes it reasonable to use on receipts, letters and screenshots with account details in them.

Not reliably. Tesseract is trained on printed and on-screen text, so neat block capitals are sometimes partly recognised, but joined-up handwriting usually comes out as garbage or nothing at all. For handwritten notes, retyping is still faster than correcting the result.

Usually the image. OCR needs sharp letters with clear contrast, roughly 20 pixels or more in height. Blurry phone photos, text shot at an angle, shadows across the page, tiny print in a large photo and busy backgrounds all cause misreads. Keep Enhance image on, crop to the text, and photograph the page straight-on in even light. Also check the language setting: reading French text as English drops the accents and confuses words.

It is Tesseract’s own average certainty across the words it recognised, from 0 to 100. Above about 85% the text is usually clean apart from the odd character; between 60% and 85% expect several errors; below 60% treat the result as a rough draft. It is an estimate, not a guarantee — a confident read can still swap 0 and O or 1 and l, so always check numbers.

No — the output is plain text. Tesseract finds blocks of text and reads them in the order it thinks is right, which works for ordinary paragraphs but can interleave the columns of a newsletter or scramble the cells of a table. Cropping each column or table region into its own image before reading gives much better order.

Before recognition it converts the image to grayscale, stretches the contrast so the darkest 1% becomes black and the lightest 1% white, and upscales images smaller than 1,400 pixels on their long side by up to 3×. That helps faded scans, grey-on-grey screenshots and small crops. It does not sharpen blur or straighten a skewed photo, and on an already crisp screenshot it can occasionally make things slightly worse — hence the toggle.