How to use
- Choose a JPG, PNG or still WebP image on your device. Check the preview to make sure the printed text is upright and large enough to read. Use the left or right 90-degree rotation controls if needed.
- Choose English for English text, or Korean + English for a document containing Korean. This selection controls recognition; it is separate from the website’s display language. Selecting Japanese as the site language, for example, does not enable Japanese OCR.
- Select Recognize text. The browser first loads the required engine and language data. When recognition finishes, correct any mistakes in the result box. Copy text and Save TXT both use the text currently visible in that editable box.
What the result means
OCR turns the visible shapes of printed characters into editable text. Unlike copying a PDF text layer, it must infer the characters from pixels, so mistakes are normal. Names, numbers, punctuation and similar-looking letters deserve a careful check before you reuse the result.
The image is processed locally. Engine files are downloaded from jsDelivr at fixed versions and English/Korean language data are loaded from this site. Those downloads do not include your image or recognized text. This page does not require an account, store a document library or call a paid recognition service.
The preview is the image that will be recognized, including your manual rotation. Large sources are reduced to at most four megapixels and a longest side of 2,400 pixels. The preview identifies a reduced image. This keeps working memory bounded but can remove fine detail; recognition is usually better when letters are clear and occupy a useful part of the image.
Examples
An upright photograph of a printed note reading “Meeting at 10:30” can become editable text. If recognition returns “Meeting at 10:3O”, change the final letter O to the digit 0 before copying. Saving TXT then includes the correction rather than the original OCR output.
For a photo containing “회의 시간 10:30”, choose Korean + English. If the photo is sideways, rotate it until the line reads normally in the preview, then recognize it. Changing the language or rotation clears the old result so you cannot accidentally copy text from a previous attempt. These examples describe the intended workflow, not a guarantee of exact recognition for every font or photo.
Limits and troubleshooting
Accepted files are at most 10 MiB and 12 megapixels, with neither dimension above 12,000 pixels. Animated PNG and WebP files, GIF, PDF and unsupported JPEG variants are rejected. File signatures and dimensions are checked before decoding; an extension alone is not sufficient. The final native image decode can still reject a damaged file.
Use even lighting, good contrast and a straight-on photo. Glare, motion blur, shadows, tiny letters and low resolution can cause missing or incorrect words. Automatic perspective correction, handwriting accuracy, table reconstruction and whole scanned PDF processing are not provided. Manual rotation fixes quarter-turn orientation, not a tilted or distorted page.
Cancel terminates the current recognition job. Selecting another image, rotating it or changing recognition language also discards an outdated result. Remove clears the source and result while retaining the language selection. A failed replacement leaves the previous valid source available. Long jobs time out so you can try a smaller or clearer image.
Frequently asked questions
Why is the result empty?
A valid image may contain no recognizable text. Check language, orientation, contrast and letter size. An empty result is shown explicitly; the tool does not fabricate words. You may type corrections into the editable result after recognition.
Can it read handwriting or a table exactly?
There is no such guarantee. The tool is intended for printed English and Korean. The TXT output contains plain text, not a spreadsheet, a faithful table layout or a styled document.
Does the website language change the OCR language?
No. The website has six display languages, while this OCR has two explicit recognition choices: English and Korean + English. Unsupported recognition languages are not implied.
Why does Copy fail sometimes?
Clipboard access can be restricted by the browser or page context. If copying fails, the result is selected so you can copy it manually. You can also save the current text as a UTF-8 TXT file.
Is the original image changed?
No. Rotation and downscaling affect the image prepared in browser memory. The original file on your device is not overwritten. Save TXT creates a separate text file.
Can I leave and return to the text later?
There is no saved document history. Copy or download the result before closing the page. A fresh session also needs access to the engine and language files; offline use is not guaranteed.