Creative Claw — xAI TTS

Read shared execution guidance once per task before using tools. It covers existing authorization, model discovery, optional cost checks, imports, and recovery.

Use this model specialist inside creativeclaw-generate-voiceover. It supplies casting and performance direction for speech/xai-tts; the outcome skill still owns the script, generation, review, and delivery.

xAI TTS is a strong fit for expressive narration, assistants, podcasts, long copy, and phone/IVR audio. Its distinctive control surface is markup inside text: square-bracket inline events create a sound or pause at one point, while angle-bracket wrapping tags change the delivery of a complete phrase.

Core workflow

1. Identify the language, audience, use case, speaker character, emotional arc, pronunciation risks, output destination, and target duration.
2. Select speech/xai-tts, then call get_model_params({ model: "speech/xai-tts" }) so runtime fields and voice availability remain authoritative.
3. Offer two fitting voices from the runtime catalog when the user wants to choose. Use Eve only when the brief gives no useful casting preference.
4. Preserve approved wording. Add supported speech tags only when they fit the requested performance.
5. Call generate_speech with model, text, and voice_id. Add language or output settings only when needed.
6. Audition pronunciation, pace, emotional fit, cue execution, tag leakage, clipping, and consistency. Revise punctuation or the smallest affected tagged span before recasting.
7. Generate separate speakers or independently editable sections in separate calls, then combine approved clips when required.

Creative Claw call shape