Eleven v4 is ElevenLabs’ text-to-speech model that it calls its most emotive yet. ElevenLabs says you can direct delivery in two ways: describe it in natural language, or insert inline audio tags such as [laughs], [light rain] or [phone buzzing]. Read the news item on the launch for what was announced. The patterns below are our own, and results vary by voice and language.
Five rules
- Say the scene, not just the words. One line of context (“a tired shop owner closing up at night”) steers tone better than a list of adjectives.
- Use audio tags sparingly. One tag per sentence at most. Too many make the delivery jumpy.
- Place a tag where the sound happens. Put [laughs] after the line it follows, not at the start of the paragraph.
- Spell hard names with phonemes. ElevenLabs says IPA phoneme support has improved, so use it for brand names and technical terms.
- Listen, then change one thing. Regenerate with a single edit so you know what helped.
Template 1: narration with a scene
Scene: a calm documentary narrator, late evening, slow and warm.
Read this text. Pause briefly after each sentence.
"The harbour was quiet. [light rain] By morning, the boats would be gone."Template 2: two-speaker dialogue
Speaker A (friendly, quick): "You tried the new recipe?"
Speaker B (embarrassed): "[laughs] I burned the first batch."
Speaker A (reassuring): "Everyone does. Try a lower heat."ElevenLabs says v4 keeps each speaker consistent across a conversation, so keep each speaker’s description the same on every line.
Template 3: voice agent greeting
Warm, upbeat, medium pace. Greet the caller, give the business name, ask how you can help. Under 15 words.For low-latency agents, ElevenLabs offers Eleven v4 Turbo. See our guide to writing voice agent prompts for the agent side.
Consent and rights
Only clone or imitate a voice when you have the speaker’s permission and the right to use it. More voice and sound examples are on our AI voice prompts page and the ElevenLabs page.