Six instrumental holiday music beds for $0.78: one Lyria call per mood

Six instrumental holiday music beds cost $0.78 on Sume, because Music Router charges a fixed $0.125 per generation and each job is rounded up to $0.13. Forty beds, one for each ad variant in a big Q4 test, cost 40 × $0.13 = $5.20.

Music is the cheapest line in a holiday ad budget by a wide margin. A single one-second slice of Omni Flash video at 720p costs the same $0.125 that buys a whole track, so the sensible question is how many moods to test, not whether to pay.

What a generation costs and what it returns

The fixed price does not change with prompt length or with optional image conditioning. sume/music-auto picks the engine, currently Lyria 3.5, and you can pin lyria-3.5 or lyria-3-pro by id from the catalog. Length comes from the prompt, such as 'a 30-second track', since the API rejects a duration field. Put exclusions in the positive prompt, for example 'instrumental, no vocals', because a non-empty negative_prompt is not supported.

Six moods, one table

Pick moods that map to buyer intent, not to genre labels. Each row is one generation.

Call it

The request is one POST. Poll the job, then read the audio artifact from result.artifacts[] where the type is audio.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: holiday-bed-flash-sale" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "Upbeat pop-funk, 112 BPM, major key. Clean bass, claps, sleigh bells on the offbeat. A 30-second track. Instrumental, no vocals."
  }'

Mixing under a voice

Timeline 1.0 accepts a soundtrack with gain_db, loop, fade_out_seconds and duck_db, so the bed can sit under a voice-over. Ducking needs a real audio spine, not silence. A one-minute render is $0.10, so a bed plus mix for one ad is $0.13 + $0.10 = $0.23.

For 40 variants that is 40 × $0.23 = $9.20 of audio work. Compare that with the cost of a 5-second clip on any current video model, and the lesson is plain: test beds aggressively, because the budget barely notices.