ElevenLabs launched Eleven v4 and low-latency Eleven v4 Turbo on 28 September. The model is built to interpret tone, pacing, emotion, character and conversational context rather than flat read-aloud. Creators can direct delivery in natural language and with inline tags such as laughs, accent notes or light ambient effects. Instant clones claim high fidelity from about 10 seconds of audio, with stronger speaker consistency across long-form and multi-speaker scenes.
What matters is the craft loop. Voiceover and character work stop being generate-and-pray. You can mark the performance beat the way you mark a script. Turbo targets agent and interactive use at roughly 100 ms median inference, while v4 aims at audiobooks, ads, games and dubbing across more than 90 languages with stronger accent adherence.
