Korean image to text gets Hangul out of pictures: a KakaoTalk conversation you want to save, a Naver blog post captured as a screenshot, a menu or street sign on a trip, a textbook page while you study for TOPIK, a product label, or a scanned school or office document. Paste or drop the image in, check the words that are underlined, and copy the text.
Korean and English are both selected on this page, because English words, brand names and numbers turn up in so much Korean text. Both are read in the same pass.
How Hangul is built
Every Hangul syllable is a block of two to four letters, called jamo, arranged in a square. 한 is ㅎ, ㅏ and ㄴ; 글 is ㄱ, ㅡ and ㄹ. The engine reads whole blocks, and the differences between blocks can be a single short stroke:
- Vowels that mirror each other: ㅏ and ㅓ, ㅗ and ㅜ, ㅐ and ㅔ
- Consonants with one extra stroke: ㄱ and ㅋ, ㄷ and ㅌ, ㅈ and ㅊ, ㅂ and ㅍ
- Final consonants (batchim): doubled ones such as ㅆ and ㄲ, and pairs such as ㄺ in 읽다, ㄵ in 앉다 or ㅄ in 없다, squeezed into the bottom of the block
At normal reading sizes, clear printed Hangul is usually read well. When text is small or blurry, these are the strokes that merge first, and a syllable turns into its near twin. Words the engine was unsure about are underlined, and a button jumps from one to the next.
Spacing (띄어쓰기)
Unlike Chinese and Japanese, Korean puts spaces between words, and those spaces matter. In print they’re often narrow, so the engine may join two words or split one. A common case is a space added before a particle: 학교 에서 instead of 학교에서, or 책 을 instead of 책을.
A spacing mistake leaves the letters intact, but it does affect search, spell-check and anything you paste into a translator. Read through the spacing as well as the letters, and use find and replace (Ctrl+F, or ⌘F on a Mac) when the same split repeats throughout a page.
Hanja and mixed scripts
Older newspapers, academic writing, legal and medical terms, and names given in brackets sometimes use Hanja (漢字). The Korean option is for Hangul, so add Chinese (Traditional) in the language menu (the globe) when your text has Hanja in it; Korean Hanja use the traditional character forms. Results on Hanja vary, especially for rare characters, so compare them with the image.
You can choose up to four languages at once, but each one makes reading slower, so add Chinese only when the page needs it.
Fonts, webtoons and handwriting
The plain print fonts used in documents, books and most apps, such as Gothic (고딕) and Myeongjo (명조) styles, read best. Handwriting-style fonts, which are popular on cafe menus, posters and social media images, are harder, and heavily decorated titles are hardest of all. Crop titles and logos out and read the body text.
Webtoons are horizontal, which helps. Crop each speech bubble on its own, since the artwork around it can confuse the engine. Sound effects drawn as part of the art won’t read as text.
Handwriting is much harder than print. Handwriting mode gives neat, carefully written notes their best chance; quick handwriting, where the parts of each syllable run together, usually needs a lot of correction. See handwriting to text for tips.
Getting a cleaner result
- Crop chat screenshots to the messages you need. Profile pictures, emoji and stickers can otherwise turn into stray characters.
- Zoom in before capturing small text. Batchim and short strokes need pixels to survive.
- Take a real screenshot rather than photographing a screen, which adds glare and blur.
- Straighten photos of signs and pages with four-corner straightening. The photo to text page has more on taking pictures that read well.
- Faded receipts printed on thermal paper often lose contrast. Raise Contrast or try Black & white before reading.
Dark-mode screenshots need nothing special, since Auto improve inverts light text on a dark background and lists each step it took. For more, read how to get accurate OCR results.
Documents and PDFs
Korean documents are often shared as PDFs, many of them exported from HWP files. If a PDF page already contains selectable text, Image to Text App takes that text directly, without OCR. Scanned pages are read with OCR one by one, and Whole document view combines them into one file. See PDF to text for more.
Where the text goes next
The result is ordinary Unicode Hangul. Copy it into KakaoTalk, a translator, Google Docs, Word or an HWP document, and it pastes like typed text. Rich copy keeps headings and lists when you paste into a document. You can also download:
- Word (.docx) for editing
- Searchable PDF, the original image with an invisible text layer you can search and select
- PDF, which for Korean opens your browser’s print dialog, where you choose “Save as PDF” so every syllable is drawn correctly
- Plain text, Markdown or HTML
Private by default
Chats and documents are often personal. Image to Text App reads them in your browser, on your device, and nothing is uploaded. History is off unless you switch it on. Read more in is online OCR private?
For Japanese or Chinese text, Japanese image to text and Chinese image to text start with the right language selected.