OCR reads the shapes of letters in an image
To a computer, the text in a screenshot or document photo is just a picture. OCR (optical character recognition) analyses those shapes and turns them into text you can copy.
Free CVT's image to text runs the open-source Tesseract engine (Apache-2.0) in your browser. It recognizes Korean, English and Spanish, and you can correct the result, copy it or save it as a TXT file.
Technical source: Tesseract.js (GitHub) · Checked 2026-10-23
Four things that decide accuracy
First, sharpness: in blurry or out-of-focus photos the strokes of letters run together and are misread. Second, size: zooming in before capturing small text makes it far easier to read.
Third, tilt and shadows: photograph documents head-on without shadows. Fourth, background: text over patterns or photos, handwriting and vertical text can produce many wrong characters. A test image of clean printed text (22 Korean and English characters) was read exactly in under a second.
Choose the language that matches the image
Pick the recognition language that matches the text in the image. For a screen that mixes Korean and English, “Korean + English” is the most accurate; for an English-only document, choosing just “English” is faster and more accurate.
Language data is downloaded once on first use (about 1.6 MB for Korean, about 3 MB for English) and then reused from your browser's storage. The recognition engine is also downloaded once, about 4 MB.
Paste screenshots straight in
After taking a screenshot, press Ctrl+V (Command+V on a Mac) on this page and the image on the clipboard is added right away, with no need to save it and pick it again.
You can also choose or drop a JPG, PNG or WEBP file. One image (up to 25 MB) is processed at a time.
Checking the result, and privacy
OCR results can mix up similar-looking characters. Check 0 and O, 1 and l, and complex Korean syllables in particular, and compare important values such as account numbers or amounts with the original.
Images and recognized text are never sent anywhere. The engine and language data are downloaded from this site, not a third-party CDN, and run in your browser, so even sensitive documents such as IDs or contracts never leave your device.
Keep the original and check the content, size and format of the result.