Prompt guide

Korean web novel translation prompt guide

A good Korean web novel translation prompt must do more than say "translate naturally." It should define the target prose, what can be localized, and what must never change.

Start with the translation role

Tell the model it is translating serial fiction for target-language readers. Ask for web novel prose, not academic translation. This helps preserve fast dialogue, scene hooks, sound effects, and short paragraphs.

Lock factual fidelity

The prompt should forbid missing, adding, or rearranging story facts. Names, actions, causality, relationship changes, foreshadowing, and reveal order must remain intact. This rule protects the story while allowing sentence-level localization.

Allow controlled localization

The prompt should allow idioms, jokes, insults, honorifics, and culturally loaded expressions to be adapted into target-language equivalents. The goal is equivalent reader effect, not mechanical word matching.

Attach memory and glossary

A prompt for episode 20 should include compact memory from earlier episodes and fixed glossary rules. Otherwise the model may choose new names, change character address, or flatten the tone of returning characters.

Ask for output only

For production translation, the model should return the translated prose only. Explanations, commentary, and evaluation tables should be separate review tasks, not part of the translated episode.

LoreLingo builds these prompt components automatically from your settings, project memory, glossary, character dictionary, and reference style.