Image to text (OCR)

Extracts text from an image (OCR — Optical Character Recognition) without sending the picture anywhere. Accepts JPEG, PNG, WebP, BMP. Supports English, Russian, and a mixed “English + Russian” mode — selectable with a switch before recognition. Under the hood it is Tesseract.js, a JavaScript port of the classic Tesseract OCR engine via WebAssembly. The engine itself (wasm + language models for English/Russian — about 25 MB total) is downloaded once on first launch and cached in the browser’s IndexedDB; subsequent recognitions are instant. Use cases: pull text out of a screenshot from a site that doesn’t expose its text as HTML (canvas-rendered JS apps and the like); read text off a photograph of a book page, an advertisement, a sign, a price tag, a timetable; grab a quote from a meme, numbers off a bank statement, a serial off a product box. After recognition the text can be copied with one click or downloaded as .txt. A “confidence” metric (~0-100%) is also shown: above 80% — almost always clean; 60-80% — worth a re-read; below 60% — Tesseract isn’t sure (likely low contrast, skewed photo, handwriting, exotic font). Accuracy depends heavily on source quality: cleanly photographed or scanned printed text recognizes near-perfectly; angled photos with glare are noticeably worse; handwriting and stylized fonts are poor. File size limit is 30 MB (beyond that browsers start running out of memory). Unlike Google Vision, Yandex OCR, ABBYY FineReader Online and similar cloud services, the image and the recognized text are never transmitted — everything happens locally on your device.