About Services Tools Blog Contact us

Instrumental Intro and Outro Beds

Generates a short instrumental piece on the local machine with the YuE music model, then cuts a matching intro and outro from the sections where the singer is silent.

  • Music Generation
  • Media production
  • Pipeline
  • Productised
  • Local/private
  • 2026

The problem

A film or an episode needs a sting that matches the rest of the piece. A second, unrelated cue from a stock library breaks the sound of the whole thing. And a sung note leaking into the intro makes it unusable under speech.

What it does

You give it a style description, a placeholder lyric that only sets the length, and the intro and outro lengths. YuE writes a score alongside the audio. The intro and outro are cut from sections where the vocal part is resting, with a margin so a sung note does not leak in. An existing backing track can be re-cut without generating it again.

Outcomes

  • The intro and the outro come from the same piece instead of two unrelated cues.
  • Cuts are taken only where the vocals rest, so speech can sit on top.
  • A new take in the same style comes from changing the starting number.
  • A cut-only path reuses a backing track that was already generated.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Estuary episode bed

Tempo
70
bpm · 4/4
Intro
16 s
bars 1–2
Outro
16 s
bars 15–16
Seed
1842
same palette
Style and length

The lyric only ends the piece. Both stings are cut from vocal rests, with a half-second margin.

Generate
Ready on this machine
Bed waveform Intro Outro Vocal rest shaded on the score
Bar 1VocalBar 16
Cut only reuses this bed

Score beside the audio

Bar length
3.43 s
4/4 at 70 bpm
Vocal-free
6
bars of rest
Margin
0.5 s
clear of sung notes
Bars
BarsVocal staffUse
1–2restIntro cut, 16 s
3restMargin before the vocal enters
4–13sungLeft unused under speech
14restMargin after the vocal leaves
15–16restOutro cut, 16 s
Dummy lyric

Open the morning and let the estuary in.

Heard by nobody. It only stops the model from running on. The cuts never include it.

Takes on disk

Render time by seed (seconds)
Output folder
SeedStyleIntroOutroStatus
1842Warm analogue, sparse drums16 s16 scurrent
1770Warm analogue, sparse drums8.0 s6.0 skept
1604Glasshouse, no percussion6.0 s8.0 skept
1908Night freight, soft pulse8.0 s10.0 scut only pending

How it works

  1. Send the style description and a placeholder lyric to YuE.
  2. Keep the score that arrives with the audio.
  3. Find the sections where the vocal part is resting.
  4. Cut the intro and the outro from those sections, with a margin.
  5. Re-cut an existing backing track without a new generation when only the lengths change.

Pipeline

flowchart LR
  Style[Style prompt] --> Song[YuE]
  Song --> Score[Score]
  Score --> Rests[Vocal rests]
  Rests --> Intro[Intro cut]
  Rests --> Outro[Outro cut]

Sample output

Sixteen second intro cut from a vocal rest in the local bed
Sixteen second outro cut from the same bed