Ad music cutdowns: 15, 30 and 60 seconds from one AI track

The cleanest way to get 15, 30 and 60 second versions of one piece of ad music on Sume is to write one prompt per length, with the same tempo, key and instruments, so each version has its own ending. The cheaper way is to generate the 60-second track once and cut shorter versions in a Timeline render, using soundtrack.fade_out_seconds for an ending. The first gives each version a composed ending. The second costs less and keeps the tracks identical for the opening seconds.

Either way the length is yours to ask for in words, because the music API has no duration field and rejects one. The rest of this post compares the two routes with Sume's own prices, shows the cut render, and says what to check before the three files go into an ad set.

Which route is better for a 15, 30 and 60 second set?

Both routes use only documented fields. Music Router generations are a fixed price: the Music 1.0 page gives $0.125 per accepted generation and says it does not vary by prompt length, and the router docs say every router model charges that same fixed Music price. A Timeline render is $0.10 per output minute, rounded up, so each cutdown render is one minute at most.

How do I keep three prompts sounding related?

Use the same musical brief and change only the length sentence and the arc. For the 15-second version, say where the one lift goes, for example at 0:08. For the 60-second version, name two moments. The docs' brief template calls for an arc with one named moment, so keep the others plain. Reusing the tempo, key and instrument lines keeps the three feeling like one family.

Each take is new, because the router has no seed, so the family resemblance comes from the prompt and not from the engine. If the three do not match closely enough, run the 15-second version twice and pick the one that sits best against the others.

A 15-second brief might read: "A 15-second track, 100 BPM, E minor. Muted guitar, tight kick, a short synth swell at 0:08, a clean stop at 0:15. Instrumental, no vocals." The 30-second one repeats the first two sentences, moves the swell to 0:16, and ends at 0:30. Keeping everything but the timing identical makes the differences between the files the ones you chose.

How do I cut the 60-second track down?

A Timeline render needs at least one video[] slot and an audio spine, even when you only want the music. Use audio.mode: "silence" with a declared duration_seconds and put the music in soundtrack. Send gain_db: 0, because Sume's compiler otherwise applies minus 16 dB to a bed that is playing alone. fade_out_seconds goes up to 10 and cannot exceed the output length.

Here is the 15-second cut of a 60-second bed over a short end-card clip. It is a vertical 1080x1920 MP4 by default. Change duration_seconds and the slot duration for the 30-second version.

curl -X POST https://api.sume.com/v1/timeline-1.0/render \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: cutdown-15-001" \
  -d '{
    "audio": {"mode": "silence", "duration_seconds": 15},
    "video": [{"source_url": "https://media.sume.com/artifacts/artf_demo/endcard.mp4",
               "start": 0, "duration": 15}],
    "soundtrack": {"url": "https://media.sume.com/artifacts/artf_demo/bed60.mp3",
                   "gain_db": 0, "fade_out_seconds": 2}
  }'

Can I check the cost before I submit?

POST /v1/timeline-1.0/plan runs the same checks without creating a job or spending anything, and returns billable_minutes and estimated_cost_usd_micros, so you can confirm the cost of a set before you submit it. It cannot predict warnings about short sources.

If you want a music-only file for a cut, there is no such render. Timeline audio split slices a Sume-hosted file by ranges and returns a wav, but it has no fade option, so a split ends abruptly unless the take already decays. Use it for stings and loops, and use the render's fade_out_seconds when the ending needs to ease out.

What should I check before shipping the set?

Listen to each file in context, under the ad, before you ship the set. A fade on a 60-second bed is gentle, but a two-second fade on a track that is still mid-phrase can sound cut off at 15 seconds. If the ending feels wrong, push the cutdown to the nearest phrase end by shortening duration_seconds, or regenerate with the arc moved earlier.

Keep the three files, their prompts and the engine named in job.request.routed_model together, so a later swap of the track is a lookup and not a hunt. The stored guide on replacing a disputed track shows how to do that across many ads at once.