Voice-overs and text: audio prompts, text prompts and text nodes
Turn a script into a voice-over, have AI draft outlines and scripts, and keep, edit and pass text along with text nodes.
Beyond images and video, Canvas works with words and sound: text nodes hold text, text prompts have AI draft outlines, scripts and more, and audio prompts read text aloud as a voice-over. Wire them together — outline → script → voice-over, for example.
Text nodes
Double-click empty space and choose Text node to write or paste ideas, outlines or scripts.
- Click the title to rename it.
- When you're not editing, the node shows its content formatted as Markdown; click the text to edit it.
- For longer text, choose Open full editor (or double-click the header) and switch between Edit and Preview. Changes save when you close it.
- Wire a text node into a text prompt, an audio prompt or a shot list to use it there.
Draft with a text prompt
Add a Text prompt and wire the text nodes you want it to work from into the card.
Write what you want — for example "Draft a short-drama outline about…" or "Expand the connected outline into a screenplay". Type @ to cite one of the wired-in text nodes; cited documents go first in the material the AI reads.
Choose the model — DeepSeek V4 Pro (the default), DeepSeek V4 Flash, Kimi K3, Claude Sonnet 5.5 or Claude Opus 5.5 — then choose Generate text.
A new text node appears below the card straight away and fills in all at once when the draft is done. Text prompts read only the text nodes wired into them, not images.
Make a voice-over
Add an Audio prompt and write the script. The voice-over reads it word for word.
To use text you already have, wire a text node into the card and type @ in the script to cite it; the document's text is inserted where you cite it.
Choose the model and voice, then choose Generate audio.
| Model | Voices |
|---|---|
| Standard voice (default) | Alloy (neutral), Echo (calm male), Fable (warm British), Onyx (deep male), Nova (bright female), Shimmer (soft female) |
| HD voice | The same voices at higher quality |
| Fish S2.1 Pro, Fish S2 Pro | Paste a voice id from Fish Audio Discovery, or a cloned voice's id; empty uses the default voice |
| Grok TTS | Ara, Eve, Leo, Rex, Sal |
A script can be up to 4,096 characters, including any text inserted by citing documents with @. Trim it if it's longer.
Voice-overs are charged per character — see AI model pricing. The estimate on the button counts only what you typed, so cited documents make the actual charge higher. Generated voice-overs are saved to your Assets library.