Singles' Day sale videos: four 15-second AI music variants

For a Singles' Day 11.11 sale, generate four different 15-second beds, one per ad variant, and render each as its own short video. On Sume that is four generations at $0.125 ($0.50) and four Timeline renders at $0.10 ($0.40), so $0.90 for four finished soundtracks, which lets you test which mood sells.

A short promo has about three seconds to land. The brief therefore front-loads energy: a clear hit at 0:00, a lift at 0:08 where the price appears, and a hard stop at 0:15 on the call to action. Change one axis per variant, for example the genre or the tempo, so a test teaches you something.

The brief to send

The first variant below is bright electronic pop. The other three keep the arc and swap the genre and instruments: a funk groove, a trap-leaning beat, and an acoustic pop clap.

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: singles-day-11-11-sale-video-music-four--001" \
  -d '{
      "model": "sume/music-auto",
      "prompt": "Energetic sale promo, 128 BPM, A minor. Punchy synth bass, claps, bright plucked lead, riser at 0:06. Hard hit at 0:00, lift at 0:08 for the price reveal, abrupt stop at 0:15. Glossy modern pop production. Instrumental, no vocals. A 15-second track."
    }'

Why each axis is set that way

Getting the length right

Fifteen seconds is a short target and the router cannot enforce it, so the prompt says it twice, once in the structure and once at the end. Check the returned file; if it is longer, cut it at the stop point in the edit, or use the render's fade_out_seconds to close the tail.

Each video is its own render, so send four requests with four Idempotency-Keys. Only the first minute of any render bills at $0.10, and 15 seconds is one started minute.

Before paying for a render, call POST /v1/timeline-1.0/plan with the same body. It is unbilled and reports the billable minutes, which for this 15-second video is 1. Then send the real POST /v1/timeline-1.0/render with an Idempotency-Key, wait up to 30 seconds in sync mode, and poll GET /v1/jobs/:id/status if it is still queued or processing. Statuses are queued, processing, completed, failed and canceled, and a 429 queue_full means try again shortly.

What it costs

Variants multiply cost linearly. The same four-variant test on 20 products would be 80 generations and 80 renders, which at the list prices above is $18.00 in total. The decision is whether the A/B signal is worth that, not whether any single clip is expensive. Scaling is simple arithmetic. One finished video here is $0.225 of audio and assembly, so ten of them are $2.25 and fifty are $11.25, before any retries. Failed jobs are not billed as accepted generations, but a regeneration you ask for is a new job, so budget one or two extra takes for a hero clip.

Check before you publish

Keep the four prompts as a table in your campaign notes. When one wins, reuse that brief as the baseline for the next sale.

Retire losing variants quickly; a flat-priced generation is cheap enough to replace.

• Name the sale event only in the on-screen text; the music brief needs mood, not a date.
• Do not state discounts or deadlines unless the shop confirmed them.
• Compare the variants by watch-through, not by which one you like best.
• Check usage terms for generated music with Sume before a paid placement.