AI music for 100 short videos: $12.50 flat on Sume Music

One hundred tracks for one hundred short videos cost $12.50 on Sume: Music generation is a fixed $0.125 per accepted generation, and the price does not vary by prompt length or by whether you add an image (read 2026-10-03). The Music Router charges the same fixed Music price for every routable model.

That makes music the cheapest line in most video budgets, and the one where the sensible strategy is to generate more than you need and pick the best. The caveat is that price is per generation, not per useful track.

What flat pricing means

The Music 1.0 docs state the fixed price directly, and the Music Router docs say every router model charges the fixed Music price per audio generation, with the catalog listing the provider list price per model for reference only. There is no duration parameter: you steer length in the prompt, for example 'a 30-second track', and the docs reject duration and duration_seconds.

So a 15-second sting and a 3-minute structured track cost the same. If your video is short, ask for a short track and trim to fit; if it is long, ask for a long one and you pay no more.

Budget for picking, not for one-shot

Because music direction is subjective, plan on several generations per video. With three candidates per video, 100 videos cost $37.50; with five, $62.50. Even then the music is a small share of the budget next to avatar or video generation, which is priced per second.

• Write the brief once and vary only tempo, key or lead instrument across candidates, using the seven-axis brief.
• Close with 'Instrumental, no vocals' for a bed under narration, as the docs recommend.
• Keep exclusions in the positive prompt: negative_prompt is unsupported when non-empty.

A request that costs $0.125

The Music Router call below is the one billable request; poll the job afterwards and read the audio artifact.

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: short-video-bed-001" \
  -d '{
    "model": "sume/music-auto",
    "prompt": "Bright upbeat pop-funk, 112 BPM, G major. Slap bass, clean guitar chops, handclaps. A 20-second track, instrumental, no vocals."
  }'

What the fixed price does not cover

The $0.125 is for an accepted generation. A request that Sume rejects for policy or validation reasons before generating is not an accepted generation, but a track you simply dislike is, and you pay for it. The docs also point out that on a policy rejection you should revise the flagged content while keeping the musical brief, and retry only within the budget you were given, rather than stripping the prompt down to a generic bed.

Nor does the music price include the rest of the pipeline. Trimming a 2-minute track to a 20-second video, mixing it under narration and burning captions are separate steps. Timeline renders cost $0.10 per output minute, a small number but one that scales with 100 videos. If you generate a track of the right length to begin with, you may skip the trim entirely.

Running the batch safely

For 100 requests, give every call its own Idempotency-Key, derived from the video id, so that a retry after a network error cannot create a second paid generation. Honour 429 responses with the retry-after header, and expect queue_full if you submit faster than your plan allows. Neither error means money was spent, but a blind retry loop without idempotency can.

Finally, record job.request.routed_model for each track. The docs say it names the engine that ran, such as Lyria 3.5 today, and if the router default changes in the future you will want to know which tracks came from which engine.

A quick budget sheet

For a 100-video series with three candidate tracks each, plan for 300 generations, or $37.50. Add a 20 percent allowance for policy rewrites and style experiments and the figure is $45.00. If each video also needs a 20-second trim through Timeline, add 100 renders at $0.10 per output minute, which is at most $10.00 if each output bills as a full minute. The total stays under $60 for the whole series, which is useful context when someone asks whether to license stock music for each video instead.