ElevenLabs drops Dubbing v2 API with emotion-matched translation

AI-dubbed videos usually share the same fatal flaw: the original emotion vanishes, replaced by a lifeless corporate narration.

ElevenLabs just launched its Dubbing v2 model via the ElevenAPI, letting developers plug end-to-end voice dubbing directly into their own products and media pipelines.

Keeping the performance, not just the words

Instead of treating dubbing as plain text-to-speech, Dubbing v2 conditions directly on the original audio performance. That means the original tone, emotional nuance, and cadence carry over across more than 90 languages.

It also brings sync-aware translation logic, automatically matching start and stop timestamps to original speech right out of the box.

Regional accents get a noticeable upgrade too, with the model distinguishing between variants like Castilian and Latin American Spanish. Background tracks, sound effects, and multi-speaker chatter are also handled much more cleanly.

The entire pipeline runs through a single API call, covering transcription, voice cloning, translation, and timing sync. Teams can also bring custom transcripts or tweak specific lines and regenerate only the edited segments.

Why it matters

We think this release moves AI localization from a parlor trick into real production infrastructure. For global video creators, game studios, and edtech platforms, producing multi-language content no longer requires assembling a fragmented chain of transcription and voiceover tools.

The real win here is segment-level regeneration. Re-rendering an entire video just to fix a single mispronounced word was an expensive nightmare, so letting editors patch individual lines makes the workflow practical.

Still, automated timing constraints mean the model may occasionally rush syllables in language pairs with massive word-count differences. It looks ready to dominate high-volume web video, though prestige film dubbing will likely keep human directors in the loop a bit longer.