Multilingual Voiceover Without the Foreign Accent

Under the homepage Languages tab, one line ("A voice is how a story is remembered") plays in 8 languages, each sounding native — because each language uses a native voice for that language. Not a coincidence; it's the only approach that works.

The accent rule

The docs put it in one sentence: language is determined by the text; accent is determined by the voice. A voice trained on English recordings reads Chinese with an English accent, and vice versa. The model does not "switch accents" on its own.

So "one voice for everything" fails the moment you go multilingual.

The right way: one character, several voices

1. Filter the community library by language and pick a native voice per target language
2. Bind them to the same character (same gender, similar age and temperament)
3. Generate each language's script separately — every clip is native-grade

That's exactly how the 8 homepage clips were made: Yun (zh), Ado (ja), Heimis (ko), Brigid (es), Nathalie (fr), Thiago (pt), Marco (de), Roberto (it) — all from the library, added in two minutes.

Common mistakes when going global

• ❌ Reading English with your own cloned Chinese voice (Chinglish)
• ❌ Generating from unreviewed machine translation (grammar errors are read verbatim)
• ❌ One stability setting for every language (Japanese and Korean usually want higher stability)
• ✅ Want your voice abroad? When cloning, upload samples recorded in the target language

One video, many languages

Workflow: source script → translate and proofread → generate per language with native voices → export subtitle timing per language. Only the audio and subtitle layers change; the picture is reused. To preserve the original speakers' voices instead, use video dubbing.