ElevenLabs released Eleven v4 (eleven_v4) today, and it replaces Eleven v3 as the default voiceover model for video ads. That covers voiceovers you generate in the video editor and the ones our AI writes for generated video ads.
In the editor
Open Voice Settings on a voiceover:
- The model toggle is now labelled Eleven v4 (it used to say V3 Model). It is on for AI-generated video ads and new audio layers; otherwise it follows the voiceover you're editing.
- Stability keeps the same three presets: Creative, Natural, and Robust.
- A Similarity slider replaces the Speaker boost switch. Eleven v4 has no speaker boost; similarity is the setting it takes instead.
- Switch the toggle off to use Multilingual v2 with its speed, stability, similarity, and style exaggeration controls, as before.
New voiceover languages
Eleven v4 speaks 11 languages that weren't available before. You can now pick them as the video language:
- Asturian
- Burmese
- Chinese (Cantonese)
- Maltese
- Māori
- Mongolian
- Occitan
- Odia
- Tajik
- Uzbek
- Yoruba
These languages need Eleven v4. If the toggle is off, Multilingual v2 shows a message asking you to turn Eleven v4 back on instead of failing. The same applies to any other language Multilingual v2 doesn't speak.
Notes
- Audio tags such as
[excited],[whisper], and[laughs]still work. Eleven v4 reads them as direction, and they stay out of the subtitles. - Japanese voiceovers on the default model now generate correctly.
- Existing voiceovers keep the audio they were made with. They only move to Eleven v4 when you regenerate them.
- Voiceover cost is unchanged at 90 credits.