Creative Claw — MiniMax Speech 2.8 HD

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. MiniMax is the strongest alternative when a native system voice, broad multilingual catalog, or global emotion/pitch controls matter more than ElevenLabs v3's line-level delivery tags.

Workflow

1. Identify the script language and regional variant, speaker profile, use case, emotion, pronunciation risks, and target pace.
2. Choose a native system voice_id from get_model_params. Use Wise_Woman only for an unspecified English brief; do not carry that default into another language.
3. Call get_model_params({ model: "speech/minimax-hd" }) and treat its voice IDs, language_boost values, and parameter ranges as authoritative.
4. Preserve approved wording. Use MiniMax pause and interjection syntax only; do not copy ElevenLabs or xAI tags.
5. Call generate_speech with model, text, voice_id, and the exact language_boost. Add the top-level emotion or speed only when useful; put pitch, volume, format, and pronunciation controls in the current runtime-supported fields.
6. Audition pronunciation, regional accent, emotion, pace, cue leakage, clipping, and loudness. Fix the smallest affected section before recasting the whole script.

Voice selection

Use get_model_params({ model: "speech/minimax-hd" }) for current system voice IDs, language labels, and provider-supported controls. Match the script language before delivery style. Offer two suitable voices when the user asks to choose. Keep one voice ID per speaker across a project.