Use speech tags in Text to Speech

Drop inline tags into the text you send to POST https://api.x.ai/v1/tts (or the streaming TTS WebSocket) for pauses, laughs, whispers, and other delivery cues. The tags stay in the request body; billing still counts the characters you send.

Inline vs wrapping

• Inline [tag] — fires at that spot (pause, laugh, breath).
• Wrapping <tag>…</tag> — changes delivery for the enclosed phrase (whisper, sing, speed).

curl -X POST https://api.x.ai/v1/tts \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "So I walked in and [pause] there it was. [laugh] I honestly could not believe it!",
    "voice_id": "eve",
    "language": "en"
  }' \
  --output expressive.mp3

curl -X POST https://api.x.ai/v1/tts \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "I need to tell you something. <whisper>It is a secret.</whisper> Pretty cool, right?",
    "voice_id": "eve",
    "language": "en"
  }' \
  --output whisper.mp3

How to write them

• Place inline tags where a person would actually pause or laugh.
• Pair tags with punctuation: "Really? [laugh] That's incredible!" lands better than a stack of bare tags.
• Use [pause] or [long-pause] when a beat needs to land.
• Wrap complete phrases, not single words.
• Combine styles when you need them: a whispered line after a pause, then normal delivery.

Preview tags in the TTS playground at console.x.ai before you ship a script.

Pitfalls

• Unary text max is 15,000 characters including tags. Longer scripts go to wss://api.x.ai/v1/tts.
• With with_timestamps: true, graph_chars includes the tag characters.
• Never call TTS from the browser with the long-lived API key. Proxy it.
• Console API credits are separate from SuperGrok's weekly pool on grok.com.