Audio AI
Audio is applied to a range on the timeline, not to a node. Drag a range with the R tool, then pick what to make.
What you can make
- Sound effects — generated from what is on screen in that range.
- Score — music across a span. Music works better under a whole scene than under a single cut.
- Voice — re-perform the dialogue in a character's voice.
- Clean up noise — reduce unwanted background.
Voices belong to characters
A voice is saved on the character in the Canvas, not on the clip. That is what keeps one person sounding the same across every shot they appear in.
For each speaker in the range you pick which character they are. A character without a voice can get one two ways: generate it from a description, or upload a short reference clip of the voice you want. Choosing None of these creates a new character with a voice.
A reference clip is a sample of timbre, not a script — a few clean seconds of that person speaking alone is enough.
What voice work does and does not do
It re-performs the line rather than converting the audio. That means you can edit the words and the direction afterwards. It also means that on close-ups the mouth may not match the new delivery exactly. Wides and off-screen lines are safe.
Limits worth knowing
- One generation carries at most three characters' reference clips. Keep a range to three speaking characters, or split it.
- More than four voices in one generation start blending together.
- Removing the original background makes the gaps between lines go silent. Keep the background unless you are replacing the whole soundscape.
Warnings for all of these appear before anything is charged.
Placing the result
Generated audio is auditioned first and placed on an audio track when you accept it. Placing does not touch the video layer.
Last updated 2026-08-25. Prices, model specs and credit costs on this page are read from the live catalogue at build time.