Voiceover and narration

Turn a written script into natural spoken audio with directed delivery: the right emotion, pacing, emphasis, and voice for a video, an ad, an audiobook, or an IVR prompt. The lever is inline [bracket] direction tags that steer how each line is read, not just the words themselves.

Inputs to collect

• The script. The exact text to speak. (Ask only if none provided.)
• The voice. A preset/voice ID, or "pick a fitting one." For a custom or cloned voice see voice-cloning.
• The delivery brief: mood, energy, pace, and audience (calm audiobook, upbeat ad, neutral IVR, urgent alert). This drives the bracket tags.
• The use: video VO, ad, audiobook chapter, IVR/phone prompt. It sets the tone and whether filler words and non-verbals belong.
• Optional: language, and whether the source text is raw LLM output that needs normalizing (numbers, dates, abbreviations).

Models

• Default expressive narration: Fish Audio S2.1 Pro (fishaudio:s2.1@pro). Flagship multilingual TTS with rich bracket-tag control over emotion and paralanguage, 80+ languages, fast streaming. Best general voiceover pick.
• Conversational / realtime delivery: Inworld Realtime TTS-2 (inworld:tts@2). Free-form natural-language steering tags and built-in settings.textNormalization for clean speech from raw text. Strong for assistants, IVR, support agents.
• Preset premium timbres: Qwen3-TTS CustomVoice (alibaba:qwen@3-tts-1.7b-customvoice). Nine preset voices with a style hint via positivePrompt. Runware Optimized.
• Streaming dialogue-style TTS: Dia2 2B (runware:dia2@2b). Real-time streaming with non-verbal cues and speaker tags. Runware Optimized.
• Confirm each is live and inspect its exact fields via runware-models + runware-run before calling. Never hardcode a stale choice.