ElevenLabs has launched Dubbing v2, an AI dubbing model designed to preserve more of the original speaker’s tone, timing and emotional delivery when translating speech into other languages.
According to ElevenLabs, the new model works directly from the original audio rather than relying only on a transcript. The company says that allows the system to carry more of the source performance into the dubbed version, including intonation, pacing and expressive detail.
The other practical feature is sync-aware translation. Instead of translating word for word and leaving editors to repair the timing later, Dubbing v2 is designed to adapt phrasing so that starts, stops and pacing stay closer to the original clip. ElevenLabs says the system supports more than 90 languages.
For localization teams, that is the part worth watching. Dubbing is not just translation; it is performance, timing, cultural adaptation, mix quality and approval. A system that gets closer to the rhythm of the original could reduce some of the mechanical cleanup involved in AI dubbing, especially for creator video, corporate media, social clips and lower-budget localization.
For higher-end film, television and streaming work, the question is not whether AI dubbing can produce a voice in another language. It is whether the result survives professional review: lip sync, emotional intent, casting suitability, pronunciation, mix integration, rights clearance and audience acceptance all still matter.
ElevenLabs is also positioning the model inside ElevenProductions, its managed localization service that combines AI dubbing with human localization work, including translators, voice casting and audio mixing. That hybrid model is likely to be the more realistic near-term route for professional media, where speed matters but unchecked synthetic voices can create quality, legal and trust problems.
The direction is clear enough: AI dubbing is moving from flat voice replacement toward performance-aware localization. The hard part is proving that emotional similarity, timing and scale can be delivered without making dubbed content feel cheap, uncanny or legally untidy.