Image to Text (OCR)

In short Optical character recognition turns the picture of a word back into the word. This runs Tesseract 5 compiled to WebAssembly, which means the recognition happens inside your browser — the image is never uploaded, never sent to an API and never stored, which matters when the thing you are reading is an invoice, a payslip or an ID document. That design has one visible cost: the first run downloads about 15 MB of engine and English language data. After that it is cached and starts immediately. Accuracy depends almost entirely on the input, and three things improve it far more than any setting. Resolution first — text smaller than roughly 20 pixels tall recognises badly, so photograph or scan closer rather than enlarging afterwards. Then contrast and straightness: crisp dark text on a plain light background at a square angle recognises well, and the same text photographed at a slant under uneven light does not. Crop to just the text you want, because surrounding graphics generate noise. What it cannot do is handwriting, decorative fonts, or text laid over a photograph, and it reads printed Latin script in English only — those are limits of this engine and language data rather than settings to adjust.