AI narration just got a lot less robotic. ElevenLabs moved its Eleven v3 text-to-speech model out of alpha and into general availability, and it’s accessible on all plan tiers, including free.
What’s New in v3
The big addition is audio tags. These are bracketed commands you insert into your text to direct how the voice performs. Tags like [whispers], [sighs], [excited], and [shouts] let you shape the emotional delivery of specific passages. If you’ve ever wished you could tell an AI narrator “this line should sound hesitant, not confident,” audio tags are ElevenLabs’ answer.
For authors producing audiobooks, this matters. A lot. Scenes with shifting emotional tones (a character moving from calm to panicked, or a tender moment interrupted by shock) can carry more nuance now without needing to generate multiple takes and splice them together.
The second major addition is the Text to Dialogue API. Instead of generating one voice at a time, you can submit a structured conversation with multiple speakers, and v3 produces a single audio file with natural transitions and emotional shifts between characters. It’s designed for exactly the kind of multi-character scenes that make audiobook production tedious.
V3 also expands language support to over 70 languages and raises the character limit to 5,000 per request. That’s roughly five minutes of audio.
The Bigger Picture
The timing on this isn’t a coincidence. On the same day, ElevenLabs announced a partnership with Bookwire, a Frankfurt-based distributor managing content for 3,500 publishers across more than 1.5 million ebooks. The focus is converting ebooks without audio versions into AI-narrated audiobooks, distributed through ElevenLabs’ Reader app.
Combine that with their existing Spotify integration and ElevenReader Publishing (which gives authors 60% on direct sales plus $0.20 per hour streamed), and the strategy is obvious. ElevenLabs is building a full audiobook pipeline, from voice generation to distribution.
What This Means for Authors
If you’ve been waiting for AI narration to get past the “sounds pretty good but weirdly flat in emotional scenes” stage, v3 is worth testing. The audio tags alone change the workflow from “generate and hope” to something closer to directing a performance.
V3 is available on all plans, but free-tier users are limited to two concurrent requests. The 5,000-character-per-request limit means you’ll still be working in chunks for full-length books. And while 70+ languages is impressive, quality varies by language, so test yours before committing to a full manuscript.
You can try v3 now at elevenlabs.io. If you already have an account, it should be in your model selection options.
Sources
- Eleven v3 is Now Generally Available — ElevenLabs’ announcement of v3 leaving alpha
- ElevenLabs Models Overview — Technical specifications and capabilities for v3
- Bookwire and ElevenLabs Partner on AI Audiobook Production — Publishers Weekly coverage of the Bookwire partnership
- What Are Eleven v3 Audio Tags — ElevenLabs’ explainer on audio tag functionality