Eleven v4

Eleven v4 is a text-to-speech model released by ElevenLabs on September 28, 2026, with the official model ID eleven_v4. It turns text into speech in more than 90 languages, accepts up to 10,000 characters per request, and supports natural multi-speaker dialogue. It is positioned for content production where audio fidelity, emotional expression, and performance direction matter.

Compared with v3, v4 improves audio quality, adherence to audio tags, cross-language delivery, and the similarity and consistency of cloned voices. It interprets the context of a whole scene and preserves speaker identity more reliably across long narration and regenerated lines. Real-time interaction uses the separate eleven_v4_turbo model; Turbo latency figures do not describe the standard model.

Key Capabilities

• Emotional Speech Synthesis: Uses textual context to shape tone, pace, and emotion for narration, advertising, or character performance.
• Fine-Grained Direction: Accepts inline audio tags and natural-language directions for local delivery and nonverbal reactions.
• Multi-Speaker Dialogue: Combines character lines into a conversation that reflects preceding turns and relationships.
• Cross-Language Speech: Supports more than 90 languages while retaining recognizable voice identity across languages.
• Voice Cloning: Supports Instant and Professional Voice Clones with improved similarity and consistency in longer content.
• Long-Form Continuity: Better preserves character voices and narrative flow across stitched requests and regenerated lines.

Technical Strengths

Feature	Benefit
Scene-Level Context	Uses relationships and preceding lines to direct responses rather than treating each sentence in isolation.
Audio Tags and Direction Prompts	Controls local emotion, accent, vocal reactions, and selected sound effects from the script.
90+ Languages	Supports consistent brand or character voices across more localization contexts.
Speaker Identity Preservation	Maintains distinctive voice traits across dialogue, long narration, and regenerated passages.
Two Voice-Cloning Paths	Instant Voice Clone supports quick creation; Professional Voice Clone addresses higher-fidelity needs.
10,000-Character Input Limit	Accommodates longer narration segments and can be combined with request stitching.

Frequently Asked Questions

Why can an Eleven v4 audio tag produce an unexpected sound effect?

The model may interpret an ambiguous tag as either a vocal direction or a request for a sound effect. Specify whether the speaker should change delivery or the scene should include a sound, then audition the relevant passage with the chosen voice; ElevenLabs notes that tag adherence is not perfect.

How do I control pauses and speed in Eleven v4?

Eleven v4 does not support SSML <break> tags, and its voice settings do not include Style or Speed sliders. Use punctuation, text structure, and audio tags to guide pacing, then listen and revise the generated result.

Which voice settings does Eleven v4 offer?

The standard controls are Stability and Similarity. Selecting a Professional Voice Clone also exposes Clone Strength, which adjusts how closely output follows the clone's training recordings; stronger adherence may leave less performance flexibility, so test it with the actual voice and script.

Does Eleven v4 keep a source voice's accent when switching languages?

When the target matches the reference voice's language, it preserves the original accent. In another language, it tends toward a natural target-language accent while keeping the speaker recognizable; have a fluent reviewer check names and regional phrasing.

How do I generate more than 10,000 characters with Eleven v4?

The 10,000-character limit applies to one text-to-speech request, so split longer scripts at natural paragraphs, scenes, or chapter boundaries. V4 improves stitching across requests, but still review joins, speaker consistency, and emotional continuity in the finished production.

Can Eleven v4 generate several speakers in one request?

Yes. Use Create dialogue or Stream dialogue and provide turns for each speaker. Use Create speech or Stream speech for single-voice narration; choosing the corresponding input structure makes speaker transitions easier to control.