| name | t2s |
| description | Text-to-speech generation. Converts a text script into a spoken-word audio file. Use for narration, voiceover, dialogue, and any plan node requiring a human voice track. |
| metadata | {"type":"primitive"} |
When to use
Use t2s when:
- The user requests narration, voiceover, or dialogue for a video.
- A plan includes a story or documentary-style sequence that needs a voice track.
- The promptIntent references a specific speaker voice, tone, or language.
Prompt conventions
The prompt for t2s is the verbatim script to be spoken, plus speaker direction.
- Script first: the exact words the voice should say.
- Voice direction (append after script): pace, tone, accent — "warm and conversational", "slow and authoritative", "excited and energetic".
- Pauses: use ellipses (...) or explicit
[pause] markers for breath breaks.
- Pronunciation guides: for unusual names or terms, add phonetic hints in brackets — "Chloé [kloh-AY]".
Model hints
- This operation is architecturally planned but the executor is not yet wired. The node will be created in the plan; execution will require a speech model integration.
Common failures
- Very long scripts (>60 seconds) may truncate. Split into multiple
t2s nodes for long-form narration.
- Emotional direction baked into the script text ("said angrily") is less reliable than explicit voice direction after the script.