Pick a target language, then dial in formality and voice — the model adapts idioms and register so it reads native, and reports the tone choices it made.