Put the drop first: rearrange a Lyria bed with one $0.01 concat job

A generated bed that builds for 10 seconds loses a Short's hook. One Timeline audio concat job reorders its sections for $0.01 on top of the $0.125 track.

4 min readSume
All posts

You can move the strongest section of a generated bed to the start of a Short by running one Timeline audio concat job on the finished track. The job costs $0.01. The track itself is one Music generation at $0.125, so the hook-first version costs $0.135 in total, with no second generation.

The idea rests on a detail of the concat API. Each entry in parts[] is { url, source_in?, duration? }, so a part is a slice of a file, not the whole file. Nothing in the Timeline audio page says two parts may not point at the same URL, so you can list one track three times with three different source_in values. The docs do not promise this use, so run it once for one cent and listen before you build a pipeline on it.

The arithmetic of a 96 BPM bar

Suppose your brief asked for 96 BPM in 4/4 and an arc with one named moment. A 4/4 bar lasts 240 / 96 = 2.5 seconds. Bars 1 to 4 are the intro (0 to 10 s), bars 5 to 8 are the drop (10 to 20 s), and bars 9 to 16 are the tail (20 to 40 s). The Music 1.0 page says the seven axes in a brief are creative directions, not guaranteed output values, so measure where the drop really lands in the file before you pick source_in numbers.

The reordered bed plays the drop for 10 seconds, then the intro for 10 seconds, then the tail for 20 seconds. The total is 40 seconds, the same as the source. A cut that falls on a bar line (10.0 s here) hides the seam better than one that falls mid-bar.

The request

Concat needs an Idempotency-Key header, parts that are already this workspace's media.sume.com audio, and the same channel layout across parts (one file repeated trivially satisfies that). Ask for wav output: the docs say wav is sample-exact, while mp3 adds priming padding at every edge, and you would have three edges.

{
  "operation": "concat",
  "output": {
    "format": "wav"
  },
  "parts": [
    {
      "url": "https://media.sume.com/artifacts/artf_demo/bed.mp3",
      "source_in": 10,
      "duration": 10
    },
    {
      "url": "https://media.sume.com/artifacts/artf_demo/bed.mp3",
      "source_in": 0,
      "duration": 10
    },
    {
      "url": "https://media.sume.com/artifacts/artf_demo/bed.mp3",
      "source_in": 20,
      "duration": 20
    }
  ]
}

What it costs against the alternatives

The table compares three ways to get a hook-first bed. Prices come from the Sume catalog.

Three routes to a bed that opens on its drop, prices as of 2026-10-08
RouteJobsArithmeticTotal
Rearrange with one concat1 music + 1 concat$0.125 + $0.01$0.135
Re-prompt for a drop at 0:001 music (one try)$0.125$0.125, but the arc is not guaranteed
Re-prompt until it works (4 tries)4 music4 x $0.125$0.50
Rearrange, then fix the seam (2 concat tries)1 music + 2 concat$0.125 + 2 x $0.01$0.145

When not to do this

Music 1.0 and the Music Router take no seed and no edit step. Google's Lyria page says generation is single-turn and that refining a clip across prompts is not supported (read 2026-10-08). A concat job is not generation, so it avoids that limit, but it cannot fix a track whose drop is weak. If the energy never arrives, rewrite the brief with an earlier named moment, for example a breakdown at 0:08 and a full return at 0:12.

Keep the output as a reusable file. The concat result gives an audio_url you can pass as the Timeline soundtrack for every market version of the same ad.

Sources

Related posts

More in Media tools

All Media tools posts

Written by Sume