Chapter 22: Audio Layers
Scene sounds (Chapter 9) attach audio to a whole scene. Audio layers put audio on the timeline instead: a layer with an in-time, a duration, a keyframable volume, and a place in the layer stack. Use them when a sound has to land on a beat rather than simply play under a scene.
An audio layer never draws anything on the canvas. It exists only in the timeline and the layer stack.
Adding an audio layer
Layer ▸ New ▸ Audio offers three items:
| Item | What it creates |
|---|---|
| Audio File… | Opens a file browser; picking a file adds a layer playing it |
| Text-to-Speech | Adds an empty TTS layer; you write the script in the sidebar |
| Volume Adjustment | A content-free layer that ducks the mix below it (see below) |
File layers
Select the layer and open the Audio section in the Layer properties:
| Field | Meaning |
|---|---|
| Source Type | File or Text-to-Speech — a layer can be switched either way |
| Source | Path or URL of the audio. May contain a {variable}, which makes it a render-time pattern |
| Content Offset | Seconds to skip into the source file before playing |
| Loop | Repeat the source for the layer’s whole duration |
| Volume | 0–200%, keyframable |
Timing lives where every other layer’s timing lives: the Timing section, or by dragging the bar in the timeline. Content Offset trims the source; in-time and duration place the layer.
Volumes above 100% are a genuine boost in the rendered movie, but browser playback caps at 100% — so a 150% layer sounds louder in the render than in editor preview.
Text-to-speech layers
A TTS layer holds a script instead of a file. In the Audio section you pick a provider (ElevenLabs or Google) and a voice, and type the text — which may contain {variables}, resolved at render time. ElevenLabs layers also pick a Model — Multilingual v2 (the default, best quality), Turbo v2.5 (balanced) or Flash v2.5 (fastest and cheapest) — per layer, so a layer that only has to be quick need not pay for the slowest model.
The generated audio is cached against everything that shapes it — the resolved script, the provider, the voice and the model — so editing the script or switching the voice or model regenerates it, and leaving them alone does not. A TTS layer inside a group is cached the same way. This is the layer-level counterpart of the voiceover editor (Chapter 17); use a TTS layer when a single line has to hit a specific moment, and the voiceover editor when you are narrating whole scenes.
Synthesis deadlines
A script with variables is spoken fresh for each viewer while the movie renders. If that takes too long, the render stops waiting and plays the recording made at publish instead, so the viewer hears the default wording rather than nothing. File ▸ Template Setup… ▸ Speech ▸ Speech Synthesis Deadlines sets how long it waits, in seconds:
| Field | Applies to | Default |
|---|---|---|
| Video render | Video, image and subtitle renders | 20 s |
| Streaming | Streamed playback, where the viewer is already watching | 10 s |
Leave a field empty to use the default. Raise a deadline for long scripts that routinely miss it; lower the streaming one if viewers should not wait for the personal version. Scripts without variables are recorded at publish and are not affected.
Volume adjustment layers
A Volume Adjustment is to audio what an adjustment layer is to video: it carries no sound, and its keyframable volume scales everything below it.
“Below it” means, in order:
- the template’s background sounds (
settings.backgroundSounds), - the scene’s own sounds,
- any audio layers — and the sound of any unmuted video layers — stacked earlier than the adjustment in the layer order.
Layers stacked above it are untouched. That ordering is what makes the classic move work:
Ducking a music bed under narration
- Put the music on the scene’s sounds, or on a low audio layer.
- Add the narration as an audio layer above it.
- Add a Volume Adjustment between them.
- Keyframe the adjustment’s volume: 100% → 30% just before the narration starts, back to 100% after it ends.
Because the adjustment sits below the narration layer, the narration stays at full level while everything under it dips — including template background sounds, which mix at movie level but still duck while this scene plays.
Outside its own in-time/duration window an adjustment layer is inert (factor 1), so a dip that should only last four seconds is a four-second adjustment layer, not a keyframe marathon.
Volume keyframes
Both audio layers and volume adjustments expose volume as a keyframable lane. Turn on auto-keyframe in the timeline and change the Volume field, or set keyframes explicitly on the layer’s volume row.
The Volume field in the sidebar always shows the animated value at the playhead, not the static value — so scrubbing through a fade shows the fade.
Audio inside groups
Grouping an audio layer keeps it playing: the mix walks into groups. Hiding a layer (or a group) removes it and everything under it from the mix.
Conditions
Audio layers and Volume Adjustments take a Visibility Condition like any other layer (see Chapter 25). While it is false the layer is not played and an adjustment does not duck; a layer in a group follows the group’s condition. A volume adjustment’s condition does not change how the template’s background sounds are ducked: they are ducked as if the adjustment always applies.
Troubleshooting
The layer is silent. Check that it is visible (a hidden layer is dropped from the mix), that the playhead is inside its in-time/duration window, that its Visibility Condition holds for the current variable values, and that no volume adjustment above it is at 0%.
Editor playback is quieter than the render. Expected for volumes over 100% — browsers cap playback at 100%.
A {variable} in the source path plays nothing in the editor.
Patterns resolve at render time. Preview against a concrete value by setting the variable in the Variables panel.
TTS keeps regenerating. The cache key is the resolved script together with the provider, voice and model. If your script contains a variable whose value keeps changing, so does the cache key.
Every TTS layer regenerated once when I opened an older template. Expected, and only once. Audio cached before voices were part of the cache key cannot say which voice made it, so it is made again the first time the template is open.
The volume adjustment ducks the wrong things. Ordering decides everything. An adjustment ducks what is below it and leaves what is above it alone, so move it up or down the stack until it sits above the sounds that should dip and below the ones that should hold their level.
← Previous: Effects and Adjustment Layers | Next: 3D Layers and the Scene Camera →