Korean Photo Translation: Getting Text Off a Photo Without Losing Northern Words

October 24, 2026 · KoryoTalk · 5 min read

Short answer: no translator "sees" a photograph — it works on text. The path from photo to translation always runs through recognition, and that is exactly where northern words are lost: the recognizer is trained on southern Korean and pulls DPRK forms toward southern ones before translation even starts — rodongja (로동자) comes back as the southern 노동자 right in the recognized text. The working order: get the text off the photo, identify the norm of the text, give the text to a tool built for that norm. For northern text that tool is the KoryoTalk bot: a dictionary base of 2.2 million Korean–Russian entries plus a separately collected layer of northern correspondences.

What people actually photograph

Behind the query "translate Korean by photo" sit four real scenarios, and the cost of an error differs:

Step 1. Get the text off the photo

Two paths, each with its own trap:

An OCR app (Google Lens, Yandex). Printed southern text it handles well. But recognition is itself a language model, and it is trained on southern Korean: forms rare in the South are recognized with an error or "corrected" into familiar ones. Northern rodongja ("worker", 로동자) can come back from OCR as the southern nodongja (노동자) — the text looks flawless while its norm has already been swapped. Same for ryeolcha ("train", 렬차) and nyeoja ("woman", 녀자).

Retype it by hand. Slower, but it distorts nothing: what is on the photo is what lands in the text. For DPRK documents, forms and handwriting this is the only path that does not lose words — and only hand-typed northern text reaches translation without the "southern smoothing".

Step 2. Identify the norm of the recognized text

Then the same check as for any Korean text: norm markers. Quick signs of the North — initial "r" and "n" (rodongja, nyeoja, ryeolcha against the southern 노동자, 여자, 기차), loanwords (komp'yutŏ, 콤퓨터, against 컴퓨터), the address dongmu ("comrade", 동무), Rosiya ("Russia", 로씨야) against 러시아. The full marker table is in the Korean-to-Russian translation guide.

The photographer's trap: if the text went through OCR, compare the recognized text with the photo. A marker can be lost at recognition — a "corrected" nodongja in the text where the photo shows rodongja is a signal that this is a northern text, not a southern one.

Step 3. Give the text to a tool built for the norm

Southern text goes to a generic translator: after clean recognition it lands in Russian passably. Northern text goes past the generic translator: munhwaŏ is absent from its data, and it will paint its southern versions on top of text already damaged by recognition.

KoryoTalk is a bot translator for the North Korean language (조선말): northern forms — rodongja, nyeoja, ryeolcha, dongmu — are the norm of the dictionary, not exceptions. One detail that matters for the photo scenario: the bot takes text (up to 4096 characters) and voice messages; it does not take photos — retype the text from the photo first, then send it. Open the KoryoTalk bot.

Continue the topic: how to translate Korean to Russian, why Google Translate doesn't do North Korean, Korean words for beginners, North Korean translator online.

Bottom line

"Translate Korean by photo" is always three steps, not one shot: text off the photo (OCR, or by hand for handwriting and DPRK documents), norm identification by markers, a tool for the norm. Printed southern text is recognized and translated automatically; northern text — after hand-typing — goes to KoryoTalk, where dongmu, Rosiya and ryeolcha are dictionary norm.

Fact check: the Korean words in this article — 동무, 로씨야, 콤퓨터, 로동자, 녀자, 렬차, 조선말, 문화어 — are confirmed against the KoryoTalk dictionary base (Oct 24, 2026). Southern forms 노동자, 여자, 기차, 컴퓨터, 러시아 appear only as marked counterexamples. Search demand (Yandex Wordstat, RU, Oct 24, 2026): "перевод по фото с корейского" 2,885/mo; "перевод с корейского на русский по фото" 2,088/mo.