Turn a script into spoken audio with a chosen voice and model — Google Gemini Flash TTS or ElevenLabs — plus language, emotion presets, and inline performance cues for nuanced delivery.
Where: canvas → left rail → Add → Audio → Gen Voice — Add an AI Voice node and connect or type a script
Key ideas
- Voice + model: Pick a TTS model (Gemini Flash TTS for Google's style-controllable voices, or ElevenLabs) and a specific voice ID; the provider is inferred from the model id.
- Language & emotion: Select a language code and an emotion preset (Happy, Sad, Whisper, Calm, …) to color the delivery.
- System prompt & inline cues: A system-prompt preset (Narrator, News Anchor, Meditation Guide, …) sets the overall persona, while bracketed inline cues (vibe / emotion / delivery) are prepended to the transcript and honored by Gemini Flash TTS and ElevenLabs v3.
- Audio output: The node outputs audio (audio_out) you can wire into an Avatar or a Video composition; results are versioned and persisted.
- Gen Voice card: VoiceGenerationNode uses the generic shared-card path. Generate speech from text with a selected model, voice, language, emotion/delivery, and speed.
- Properties sidebar: renderAudioNodes.tsx
- Opened surfaces: Audio preview
Steps
- Add a Voice Generator node and type your script (or connect a Text node to its prompt).
- Pick a model (e.g. Gemini 3.1 Flash TTS) and choose a voice.
- Set the language and an emotion preset, and optionally a system-prompt persona.
- Add inline cues (vibe / emotion / delivery) for nuanced takes if desired.
- Click Generate; preview the audio, then download or wire audio_out downstream.
Card and inspector fields
- Model and voice.
- Language and system prompt.
- Emotion/delivery controls.
- Speed from 0.5× to 2× in the shared UI; provider routes may clamp or convert it to a delivery instruction.
Opened surface
Audition every version in the audio preview before using it in Lip Sync, Digital Twin, Audio Render, or Video Composition.
Gen Voice controls and presets
- Add path: Flowgen → Add → Sound → Gen Voice.
- Card: VoiceGenerationNode.
- Properties: renderAudioNodes.tsx.
- Controls: Model; voice; language; system prompt; emotion; speed; versions.
- Opened surfaces: Audio preview.
FAQ
What do the [bracketed] cues do?
They are inline performance tags prepended to your transcript. Gemini Flash TTS and ElevenLabs v3 honor them (e.g. 'energetic yet friendly', 'whispering'); older models simply ignore them, so they're always safe to add.