ElevenLabs API Guide: Getting Started for Developers

Authentication and Setup

ElevenLabs API uses API key authentication. Generate your key in the Profile Settings section of the ElevenLabs dashboard. Include it in all API requests as an xi-api-key header. Key format: a long alphanumeric string that you should treat as a secret credential — don't commit it to public repositories or expose it in client-side code.

Official SDKs are available for Python and JavaScript/TypeScript from ElevenLabs' GitHub repository. For Python: pip install elevenlabs. For JavaScript: npm install elevenlabs. The SDKs wrap the REST API with typed interfaces and handle authentication and streaming automatically.

Core API Endpoints

The text-to-speech endpoint is the one you'll use most: POST /v1/text-to-speech/{voice_id}. The request body includes text (your input script), model_id (the generation model — use "eleven_multilingual_v2" for best quality), and voice_settings (stability, similarity_boost, style values). The response is binary audio data in your requested format.

The streaming endpoint at /v1/text-to-speech/{voice_id}/stream uses identical parameters and returns a chunked streaming response. Use this for real-time applications where first-audio latency matters. The voices list endpoint (GET /v1/voices) returns all available voices including your custom cloned voices, which is useful for letting users select voices programmatically in your application.

Rate Limits by Plan

Free plan: 2 concurrent requests, limited RPM — unsuitable for production. Starter ($5/mo): 5 concurrent requests, significantly higher RPM. Creator ($22/mo): suitable for small-scale production applications with moderate concurrent users. Pro ($99/mo): comfortable for mid-scale applications. Scale ($330/mo): designed for high-throughput production with dedicated limits.

All plans draw from the same monthly character pool whether usage comes via API or the web interface. There's no separate API-only credit pool — your API calls consume from your plan's monthly allowance at the same rate as web interface generation.

Modelling API Costs at Scale

For applications where voice generation volume scales with user behaviour, character consumption can grow quickly in ways that are hard to predict. A safe approach: calculate your expected average characters per API call, estimate your daily API call volume at expected user load, and multiply by 30 to get monthly character consumption. Compare to your plan allowance and add a 20% buffer.

For applications with unpredictable usage patterns, PlayHT's unlimited API plan may offer better financial predictability than ElevenLabs' character-based model. The right choice depends on whether voice quality or cost predictability is the higher priority for your application.