Image to Text applies OCR (Optical Character Recognition) to any single image — a photo, a screenshot, a downloaded graphic, a scanned page saved as JPG or PNG — and turns whatever text it contains into plain, editable characters you can copy, correct, and reuse. This is the general-purpose version of OCR on this site: it isn't built around a camera-capture workflow like Scan to Text, and it doesn't reconstruct a searchable copy of an existing PDF the way PDF OCR Search does. It exists for the simpler, more common case — you already have an image file sitting on your device, and you need the words inside it as actual text rather than as pixels.
OCR itself works by pattern recognition rather than "reading" in any human sense. The image is first pre-processed — converted to grayscale, thresholded to separate dark text from light background, and often deskewed if the text is slightly rotated — and then segmented into lines, words, and finally individual glyphs. Each glyph shape is compared against a trained model of what characters in a given language typically look like, and the engine assembles its best guess at the underlying text, word by word, using dictionary and language-frequency data to resolve ambiguous shapes (distinguishing a lowercase "l" from a numeral "1" from a capital "I", for instance, is a classic OCR challenge that context and dictionary lookups help resolve). This tool runs Tesseract.js, a WebAssembly build of Tesseract — the open-source OCR engine originally developed at HP in the 1980s, open-sourced in 2005, and now maintained with Google's backing. Tesseract is widely regarded as the most capable open-source OCR engine available and is the same underlying technology behind a large share of commercial and open-source OCR products.
Accuracy in practice depends heavily on image quality rather than on the engine itself. Clean, high-contrast printed text — a screenshot of an article, a crisp photo of a printed page, a well-lit sign — routinely gets recognized with well over 95% character accuracy. Accuracy drops noticeably with low resolution (anything under roughly 150-200 effective DPI starts to blur glyph edges together), poor lighting or shadows across the text, skewed or rotated capture angles, unusual or decorative fonts, and especially handwriting, which Tesseract was never primarily designed to handle. Understanding that trade-off is more useful than treating OCR as a black box: if a result comes back garbled, the fix is almost always to retake or re-crop the source image rather than to expect the engine to compensate.
Because the entire recognition pipeline — image decoding, preprocessing, character segmentation, and text assembly — runs as WebAssembly inside your own browser tab, no image you upload here is ever transmitted anywhere. That matters for the kind of content people actually run through OCR: screenshots of private conversations, photos of ID documents or forms, internal business receipts, or personal notes. The recognized text lands in an editable text area on the page itself, so you can fix any misread characters before copying it to your clipboard or saving it as a plain .txt file.
Capturing the text live with your camera instead of a saved file? Scan to Text is built for that. Working from a whole PDF? Searchable PDF handles multi-page documents.
