Inline Audio Tags as director notes
Write [whispers], [excited], [laughs], [sighs], [sarcastically] in copy. For narration and ad reads that need specific emotion, less “generate again and hope.”
ElevenLabs · ElevenLabs v3
ElevenLabs v3 — Direct emotion with inline tags
ElevenLabs GA’s most expressive TTS: inline Audio Tags, multi-speaker dialogue, 70+ languages; 4.9% error rate. Default for narration, dialogue, and performance reads.
- Inline Audio Tags
- Multi-speaker dialogue
- 70+ languages
- 4.9% error rate (GA)
What changed in ElevenLabs v3 GA
GA vs Alpha: 72% user preference; digit, symbol, and notation error rate 15.3%→4.9% — fewer retakes when voiceover misreads amounts and dates.
Voiceover problems ElevenLabs v3 solves
Inline Audio Tags for emotion, multi-speaker dialogue, 70+ languages in one workflow — when narration feels flat, characters talk over each other, or accents drift across languages, fix tags and speakers first instead of blind rerolls.
Multi-speaker dialogue
Text to Dialogue handles multi-character turns, interruptions, and emotional handoffs. Podcasts, game cutscenes, story demos — fewer “one voice playing two roles.”
70+ languages, one workflow
Global content without swapping models per locale. Localized VO and multilingual ads: align specs, then edit copy and tags — not reopen a whole pipeline.
One model, three production workflows
Narration to dialogue to ad reads — different deliverables, same failure cost: rerecords and missed deadlines.
Audiobook / documentary narration
Use tags for breath and emotional arc when needed. Flat machine read or weak climax breaks immersion; ElevenLabs v3 defaults to narrative expressiveness.
Game / story multi-character dialogue
Multi-speaker shared context. Characters talking over each other or wrong emotion kills cutscenes; dialogue mode beats stitching single-speaker TTS.
Ad reads and product demos
Name excitement, restraint, irony with tags. Wrong read tone means rerecording the whole spot; set tags first, listen, less blind sampling.
Three steps to run ElevenLabs v3
- 01
Open the iMini audio workspace
Go to Audio and select ElevenLabs v3.
- 02
Write copy in the official structure
Body copy + lowercase English bracket tags, e.g. [whispers] Something’s coming… [sighs]. Multi-character: use dialogue flow to split speakers.
- 03
Generate, listen, adjust tags
Listen for emotion and diction first; if off, change tags/wording and regenerate — don’t rewrite the whole script blindly.
Using ElevenLabs v3 on iMini
Align before you start: free credits, model name, tag syntax, when to default to it.
Generate ElevenLabs v3 free on iMini
Audio Tags · multi-speaker · 70+ languages — alongside image/video models, open and go.
Use ElevenLabs v3 freeNo ElevenLabs API key required.