Summary
OpenRouter recently added text-to-speech models that could serve as TTS providers alongside ElevenLabs:
- GPT-4o Mini TTS — $0.60/1M chars, 4K context, OpenAI voices
- Gemini 3.1 Flash TTS — $1/M input, $20/M output tokens, 8K context, 200+ inline audio tags (e.g.
[whispers], [laughs], [excited]), 70+ languages, multi-speaker, SynthID watermarking
- Voxtral Mini TTS — $16/1M chars, zero-shot voice cloning
Currently, OpenClaw's TTS subsystem only supports elevenlabs and sag as provider IDs. The OpenRouter TTS API uses a different request/response format than ElevenLabs, so it cannot be used as a drop-in replacement.
Why this matters
- Cost flexibility: OpenRouter offers pay-per-use TTS without a monthly subscription (unlike ElevenLabs's $22/mo Creator plan)
- Audio tags parity: Gemini 3.1 Flash TTS supports 200+ inline audio tags, very similar to ElevenLabs v3's audio tag system (
[laughs], [whispers], etc.) — this is a natural fit for OpenClaw's [[tts:...]] directive
- Multi-speaker: Gemini 3.1 Flash TTS supports 2 independent speakers with per-speaker voice/style config
- Language coverage: 70+ languages vs ElevenLabs's ~30
- Fallback: Users already using OpenRouter for LLMs could use the same API key for TTS
Suggested implementation
Add openrouter as a supported TTS provider ID that calls the OpenRouter /api/v1/models/{model}/speech endpoint (or equivalent) with the configured model ID.
Config would look like:
{
"messages": {
"tts": {
"provider": "openrouter",
"providers": {
"openrouter": {
"apiKey": "...",
"modelId": "google/gemini-3.1-flash-tts-preview",
"voiceId": "...",
"languageCode": "de"
}
}
}
}
}This would allow existing OpenRouter users to add TTS without a separate ElevenLabs subscription.
Environment
- OpenClaw 2026.4.22
- Currently using ElevenLabs v3 with audio tags (working well)
- OpenRouter API key already configured for LLM fallbacks
Summary
OpenRouter recently added text-to-speech models that could serve as TTS providers alongside ElevenLabs:
[whispers],[laughs],[excited]), 70+ languages, multi-speaker, SynthID watermarkingCurrently, OpenClaw's TTS subsystem only supports
elevenlabsandsagas provider IDs. The OpenRouter TTS API uses a different request/response format than ElevenLabs, so it cannot be used as a drop-in replacement.Why this matters
[laughs],[whispers], etc.) — this is a natural fit for OpenClaw's[[tts:...]]directiveSuggested implementation
Add
openrouteras a supported TTS provider ID that calls the OpenRouter/api/v1/models/{model}/speechendpoint (or equivalent) with the configured model ID.Config would look like:
{ "messages": { "tts": { "provider": "openrouter", "providers": { "openrouter": { "apiKey": "...", "modelId": "google/gemini-3.1-flash-tts-preview", "voiceId": "...", "languageCode": "de" } } } } }This would allow existing OpenRouter users to add TTS without a separate ElevenLabs subscription.
Environment