Duck a Suno or ElevenLabs track under a voice-over in Sume Timeline

If you made a track in Suno or ElevenLabs and need it to sit under a voice-over, Sume Timeline 1.0 can mix it in one render. The soundtrack field takes a bed with url, gain_db, loop, fade_out_seconds up to 10 and duck_db from 0 to 20, and the bed ducks under a real audio spine. The render is $0.10 per output minute.

The tools cover different steps. Suno and ElevenLabs make the music; Sume Timeline places it and mixes it with speech and video.

The soundtrack block

Every URL in a Timeline render must already be an artifact or asset in your workspace on media.sume.com. That means an outside track goes through a media import first. The bed field then does the mixing: gain_db sets level, loop repeats a short bed for a long spine, fade_out_seconds ends it smoothly, and duck_db lowers the bed when the spine carries sound. Ducking needs a real spine; it does not work when audio.mode is silence.

{
  "audio": {"url": "https://media.sume.com/artifacts/artf_demo/voice.wav", "duration_seconds": 45},
  "soundtrack": {
    "url": "https://media.sume.com/artifacts/artf_demo/bed.mp3",
    "gain_db": -12,
    "loop": true,
    "fade_out_seconds": 3,
    "duck_db": 10
  },
  "video": [
    {"source_url": "https://media.sume.com/artifacts/artf_demo/shot.mp4", "start": 0, "duration": 45}
  ]
}

Cost of the finished 45 seconds

A 45-second render rounds up to one billable minute, so Timeline is $0.10. If the bed comes from Sume Music Router, add $0.125 per accepted generation: $0.10 + $0.125 = $0.225 before the voice and video. If the bed comes from a Suno or ElevenLabs plan, the track already sits inside that subscription, so the Sume addition is the $0.10 render.

Send POST /v1/timeline-1.0/plan first if you want the estimate without creating a job. It does not predict warnings from short sources being padded or looped.

Ducking is easiest to judge by ear, so render a short test first. A 10-second render rounds up to one billable minute and costs $0.10, which is a cheap way to try duck_db at 6, 10 and 14. Because the plan endpoint is unbilled, you can also confirm the billable minutes for the full-length version before you commit. When the bed is longer than the spine, the output length is still set by audio.duration_seconds, so a long track does not extend the video. If the bed is shorter and loop is false, it simply ends, which is a useful way to leave the last seconds clear for the voice.

• Keep the bed shorter than the spine only if you set loop to true.
• Choose duck_db by ear; 0 means no ducking, and the doc range goes to 20.
• Send an Idempotency-Key with each render so a retry does not double-charge.

Rights come first

The mix is mechanical; whether you may use a track in an ad is a rights question. Suno's page points to its terms for limits on commercial use, and ElevenLabs lists a commercial license from Starter upward. Keep the source, plan and date with the project file.