Eleven v3 alpha

Eleven v3 (alpha) advances ElevenLabs’ stack with more expressive prosody, tighter timing, and better voice consistency across languages. You can synthesize from text or drive delivery from a guide recording, then keep the same voice while translating into other languages with natural phrasing and matching cadence. The model responds to style prompts and reference takes, supports controllable speaking rate and energy, and returns stable word- or sentence-level timestamps so edits land precisely in a timeline. Latency is trimmed for live agents and interactive apps, while longer batches render with cleaner diction and fewer artifacts for audiobooks, training content, and localization. Voice cloning and safety controls remain central—captured voices can be restricted to approved projects, and outputs can be shaped to brand tone without sacrificing intelligibility. As an alpha release, features and voicing range are still expanding, but v3 already makes day-to-day production faster: write or translate, audition variations, and publish audio that sounds intentional rather than synthetic.