ElevenLabs has launched Dubbing v2, an update to its AI dubbing technology that supports more than 90 languages and accents while aiming to preserve the original performance's tone, pacing, delivery and emotional intent. Its central change is a sync-aware translation system designed to adapt wording for natural speech without losing alignment with the source audio.
For global video localization, that combination addresses a practical constraint that has long shaped dubbing workflows. A translation can be linguistically accurate but still sound unnatural when spoken, or it can run too long or too short for the original scene. Dubbing v2 is intended to handle both issues by conditioning on the original performance rather than treating a transcript as the sole input.
According to ElevenLabs' official Dubbing v2 announcement, the system automatically aligns starts, stops and pacing with the source content. The company positions the release as a way to retain a speaker's personality across languages, rather than simply generating translated voice tracks.
What Dubbing v2 changes
The important distinction in ElevenLabs' description is performance conditioning. Instead of working only from written dialogue, Dubbing v2 uses characteristics of the original delivery to inform the generated dub. That matters because spoken communication carries meaning through timing, emphasis and emotion as well as words.
