Multi-speaker dialogue audio API

ElevenLabs Text-to-Dialogue v3 / v4 API

Turn scripted dialogue into natural multi-voice audio.

The ElevenLabs Text-to-Dialogue v3 / v4 API converts a sequence of dialogue turns — each with its own voice — into a single expressive audio file. Built for conversations, it captures emotion, pacing, and turn-taking across multiple languages.

Read the official model and pricing documentation
Try playground
up to 10
Voices per request
up to 2,000
Characters
MP3
Output
ElevenLabs
elevenlabs-text-to-dialogue-v3async task
up to 10
Voices per request
up to 2,000
Characters
MP3
Output

Model capabilities

Built for production API workflows, not one-off demos.

Use ElevenLabs Text-to-Dialogue v3 / v4 from the same platform surface as the rest of your video and image stack: API keys, credits, logs, webhooks, and docs stay consistent across providers.

Per-line voice control

Assign a distinct ElevenLabs voice to each dialogue turn for believable multi-character conversations.

Expressive delivery

Eleven v3 and v4 capture emotion, emphasis, and natural turn-taking for lifelike back-and-forth dialogue.

Multilingual

Auto-detect or pin the language across supported languages, with consistent voices throughout.

Stability control

v3 supports stability 0, 0.5, or 1; v4 supports any value from 0 to 1. Both default to 0.5.

API workflow

Submit tasks, track progress, and return generated assets.

01

Submit a dialogue task

Send a dialogue array of { text, voice } turns, optional stability and language_code, and a callback URL to the audio endpoint.

02

Track async progress

Use the returned task ID to poll status, or let the callback deliver completion and failure events.

03

Receive hosted audio

Completed tasks return a hosted MP3 URL through the same response shape used by other models.

POST /v1/audio/generations
curl -X POST https://api.aivideoapi.ai/v1/audio/generations \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "elevenlabs-text-to-dialogue-v3",
    "input": {
      "dialogue": [
        { "text": "Hey! Did you finish the report?", "voice": "TX3LPaxmHKxFdv7VOQHJ" },
        { "text": "Almost — just adding the final numbers.", "voice": "nPczCjzI2devNBz1zQrb" }
      ],
      "stability": 0.5,
      "language_code": "en"
    }
  }'
What teams build
Podcast and audio drama dialogue
Game and NPC conversations
Audiobook character voices
Language-learning dialogues
IVR and voice agent scripts
Explainer and ad voiceovers

Frequently asked questions

Answers about the ElevenLabs Text-to-Dialogue v3 / v4 API.

What is the ElevenLabs Text-to-Dialogue v3 / v4 API?

It is a multi-speaker text-to-speech API powered by ElevenLabs Eleven v3 and v4, exposed through AI Video API. You provide an ordered list of dialogue turns, each with its own voice, and receive a single hosted audio file rendering the full conversation. Generation is asynchronous with callbacks or polling.

How is the Text-to-Dialogue v3 / v4 API priced?

Both models cost 24 credits per 1,000 Unicode characters, with no minimum charge. Count text across all turns, including spaces and audio tags, and round the base total to two decimals before applying your account multiplier. Credits are pre-charged on submit and refunded on generation failure.

How many voices and characters can one request use?

A single request supports up to 10 distinct voices and a combined total of 2,000 Unicode characters across all dialogue turns. Each turn specifies its own ElevenLabs voice ID.

Which languages are supported?

Eleven v3 supports multiple languages. Omit language_code to auto-detect, or pin a specific code such as en, zh, ja, or fr.

Pricing and usage

Clear model options with shared credits.

Per 1,000 characters
24 credits

Unicode characters; no minimum charge

Voices per request
up to 10

Distinct voice IDs

Character limit
2000

Total across all turns

Start building with ElevenLabs Text-to-Dialogue v3 / v4 in AI Video API.

Create one API key, use one credit balance, and switch between video and image models without provider-specific plumbing.

Read docs