Audio
Speech, music, sound effects, and audio file tools.
Audio generation creates speech, music, or sound effect files on the canvas. You can also transcribe or clean existing audio files.
From text or note nodes
Open the generate bar with Generate audio on a text or .md note node.
Intent and mode labels
| Intent in bar | Mode label | Needs |
|---|---|---|
| Speech | Text to Speech | Prompt text; optional voice source |
| Music | Music | Prompt text; MiniMax With Lyrics mode needs upstream lyrics text |
| Sound | Sound Effect | Prompt text describing the sound |
Voice sources for TTS
On models that support voice cloning or design:
- Built-in — preset voices from the model.
- Reference — upstream audio arrow for voice clone. Run disables if no audio is connected.
- Design — upstream text arrow for a voice design prompt. Run disables if no text is connected.
Hover the disabled Run button to read the exact requirement.
BreezeBlue voice references
BreezeBlue creates short audio candidates to use as voice references in a video. You can keep the existing description-and-script workflow.
- Connect BreezeBlue in plugin settings and enable Breeze TTS 2 Multilingual.
- Write a voice description in a text node, then connect it to a separate node containing a short spoken line.
- Select Speech and Design on the script node, choose the Breeze model, and set Candidates to 1, 2, or 3.
- Click Run. Each candidate appears as a separate audio node.
- Listen to the candidates and connect your chosen audio to a video prompt whose model supports audio references.
The description and spoken line must each contain 3–500 characters. This workflow saves the preview audio in your vault; it does not require saving a reusable voice or generating another speech recording. Language selection is handled by the multilingual model. Generating more candidates consumes more preview credits.
From audio file nodes
Select an audio file node (mp3, wav, flac, m4a, ogg, aac):
| Button | Result |
|---|---|
| Speech to text | A text node with transcription |
| Isolate audio | A cleaned audio file node |
Both also show Download and Mark.
Audio models
Enable models in Settings → Models → Audio Models. Registered audio models include:
- DashScope CosyVoice 3.5 Plus
- DashScope CosyVoice 3.5 Flash
- DashScope Qwen Voice
- ElevenLabs TTS v3
- MiniMax TTS
- Grok TTS
- ElevenLabs Music
- MiniMax Music
- ElevenLabs Sound Effects v2
Each model exposes its own voices, duration controls, and other parameters in the generate bar.