> For a complete page index, fetch https://docs.synthflow.ai/llms.txt. For full documentation content, fetch https://docs.synthflow.ai/llms-full.txt.

# ElevenLabs

> Connect ElevenLabs for custom voices in the agent editor and choose from ElevenLabs voice models for text-to-speech.

![Synthflow AI integration with ElevenLabs](https://storage.googleapis.com/granular-changelog/doc-images/elevenlabs_cover.svg)

[ElevenLabs](https://elevenlabs.io) provides voice cloning and text-to-speech for Synthflow agents. You can use Synthflow's built-in ElevenLabs access or connect your own API key when you need voices from your ElevenLabs account.

## Custom voices with your API key

When you connect your ElevenLabs API key, your custom voices appear in the [voice selector](/voice-configuration). Choose the speaker there, then pick the voice model that should synthesize speech for that agent.

> **Warning**
>
> Be aware that the synthesizer models available in Synthflow may not be enabled on your ElevenLabs account. When a model is missing from your plan, you may notice lower voice quality or higher latency while using your own API key.

If you only need a different built-in speaker, you can use Synthflow's default ElevenLabs access without connecting a key. Contact Synthflow support if you need a voice added to the shared library.

## ElevenLabs voice models

In [Voice configuration](/voice-configuration), the **Voice Model** dropdown lists these ElevenLabs engines (alongside [Synthflow TTS](/voice-configuration#synthflow-tts)):

| Voice model                    | Best for                                                                                                                                                                                                                                           |
| :----------------------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **ElevenLabs Turbo v2**        | English only. Extremely low latency for English agents.                                                                                                                                                                                            |
| **ElevenLabs Turbo v2.5**      | Low latency across 32 languages when speed and non-English support both matter.                                                                                                                                                                    |
| **ElevenLabs Flash v2**        | Recommended for English. Ultra-fast real-time synthesis (about 75 ms) for English calls.                                                                                                                                                           |
| **ElevenLabs Flash v2.5**      | Recommended for non-English. Ultra-fast real-time synthesis (about 75 ms) for multilingual calls.                                                                                                                                                  |
| **ElevenLabs Multilingual v2** | Natural, expressive speech across 29 languages when latency is less critical than quality.                                                                                                                                                         |
| **ElevenLabs V3**              | Beta. The latest ElevenLabs model, with improved naturalness. Use when you need [audio tags](/voice-configuration#audio-tags) in prompts for accent, emotion, and delivery control. Higher latency than Turbo or Flash, so test before production. |

> Custom voices and ElevenLabs text-to-speech models for your agents