Pull text out of photos and screenshots. Korean and English are read together, and every line carries its own confidence score.
How to use it
- Upload images containing text. Several at once is fine.
- Choose the language. For a document with no English in it, “Korean only” makes fewer mistakes.
- Press “Read text”. The first run takes a few extra seconds to download the recognition model.
- Copy the result or download it as .txt. Lines shown faded scored low — check those.
Is my image uploaded?
No. The recognition engine and language data are downloaded to your browser and everything happens on your device. That makes business cards, ID documents and contracts safe to run through it.
Why is the first run slow?
It downloads the engine (about 2.7 MB) and the language data (1.6 MB for Korean, 3.9 MB for English) once. After that they stay in your browser cache and recognition starts immediately.
Does it read handwriting?
Barely. The engine is trained on printed characters, so handwriting accuracy drops sharply. Use it for print, screenshots and signage.
How do I improve accuracy?
Bigger, sharper characters help most. Straighten a tilted photo first with the Crop tool. If the resolution is very low, re-shooting beats upscaling. Narrowing the language to what the document actually contains also makes a real difference.
Is “Korean only” always more accurate?
No. It is better for documents that are purely Korean, but worse the moment Latin text appears. On the same test image, “Korean + English” misread 서울시 as “MBA” yet got the email address right, while “Korean only” fixed 서울시 and turned the email address into unreadable characters. Match the setting to what the document actually contains.
Is table or document formatting preserved?
No. Only the text is extracted, line by line. Table structure, fonts and colours are not carried over.
Can I extract text from a PDF?
Uploading a PDF converts its first page to an image and reads that. If the PDF already contains real text, copying it from a PDF viewer is more accurate. OCR is for scans, where the text exists only as a picture.