Whether the dialogue can be kept without the music
The usual objection to generated audio is the music rather than the speech. One entry documents a mode returning speech alone; another scores the episode after the shots exist and has a switch that returns the dialogue without it. Everywhere else the tracks arrive mixed into one file. From 4 source pages, read 2026-09-12 to 2026-10-13; no maker pays to be listed.
| Model | What is documented | Detail |
|---|---|---|
| SceneMixer | A score composed per episode and mixed under the voices | A Music switch returns the dialogue-enhanced version |
| sync-3 | Whatever was supplied | Separation is upstream, not here |
| Vidu | Speech, effects and music, with a speech-only mode | On by default in the API |
| Wan 3.0 | One track, on unless switched off | No separation documented |
Inclusion rule. Entries whose documentation describes what the generated audio track contains, or a mode that changes it. An entry silent about the contents of its audio does not earn a row. Order. Alphabetical by model name.
1Music is a decision that belongs to an editor
A generated bed arrives with a creative choice already made, and it is a choice nobody asked for. For a single clip that is convenient; across a series it means the score is whatever the prompt produced on each shot, which is the opposite of how a score works.
A documented way to keep the dialogue and drop the rest removes the objection without removing the feature. Two entries publish one. One is a speech-only mode on the model whose default is the fullest mix. The other moves the score out of the shot altogether: music is planned for the whole episode, composed and mixed under the voices after the shots exist, with a switch beside the player that turns it off.
The second arrangement answers the objection in the first paragraph directly. A score planned across an episode can stay coherent through a cut in a way a bed generated shot by shot cannot.
2Baked-in audio makes a mute impossible
Where the tracks are mixed into the returned file, an unwanted bed cannot be muted in the edit; the shot has to be generated again with different instructions. That converts a small aesthetic objection into a cost and a delay.
It also means a scene meant to play dry needs the prompt to say so. One vendor documents the phrasing for asking for silence, which is unusual and worth copying: an explicit negative rather than an absence.
3Supplied-audio products never have the problem
Where the sound is an input, whatever was handed over is what plays, so separation happened upstream in a mix a production controls. That is a real advantage of the audio-driven arrangement and it is rarely stated as one.
The cost, as always with that arrangement, is that the sound has to exist first. A team with a mix already has the cleanest path; a team hoping to hear something within the hour does not.
- Audio sourceQ3 outputs speech, sound effects and music natively, on by default in the API, with a speech-only modethree tracks in one pass
- Audio sourceThe audio parameter defaults to true so the returned video carries an audio track, and false returns one withouton unless it is switched off
- Audio sourceThe accepted pairs are video with audio, video with text, image with audio and image with textsupplied or read from text
- Where the music comes fromEach episode is scored once its first shot exists: music planned shot by shot, composed and mixed under the voices, at no extra charge
- Keeping the dialogue without the musicA Music switch beside the player gives the dialogue-enhanced version without the score; there is no track picker and no volume control
4Sources
Every line is read from the vendor documentation linked on the speech controls page, checked 2026-09-12. What counts as documented is on how read. Nearby: Regenerating a shot, Consent in the docs, A second control.