AI music for a 15-second bumper ad: per-minute vs flat pricing

For a 15-second bumper ad, flat per-track pricing is the more expensive option. ElevenLabs lists its Music API at $0.15 per minute, so 15 seconds is about $0.04. Google lists lyria-3.5 at $0.08 per song and a 30-second clip model at $0.04. Sume's Music Router charges a fixed $0.125 per generation. Sume wins on long beds and on the surrounding Timeline workflow, not on the shortest cues.

The numbers

Prices are from the ElevenLabs API pricing page and the Gemini API pricing page, plus the Music 1.0 page for Sume, all read on 2026-10-04.

Why the gap exists, and when it flips

A flat fee is the same at 15 seconds and at three minutes, so it is poor value at the short end and good value at the long end. At $0.15 a minute, ElevenLabs reaches $0.125 at about 50 seconds. Beyond that, Sume's flat price is lower: a 2-minute bed is $0.30 on ElevenLabs and $0.125 on Sume. Google's $0.08 stays below Sume's flat fee at every length, but you call Google directly.

• Under about 50 seconds: per-minute pricing is cheaper than Sume's flat fee.
• Over about 50 seconds: Sume's flat fee is cheaper than ElevenLabs' per-minute rate.
• Versus Google direct: a lower list price, but none of Sume's jobs, hosted artifacts or Timeline.

Making a short cue on Sume

Sume rejects duration, so write "a 15-second track" in the prompt, or generate a longer bed and cut it. Timeline audio split slices a range from a file for $0.01 per job, and a Timeline 1.0 render bills $0.10 per output minute rounded up, so a bumper render is one billable minute. One generation can feed 15-, 30- and 60-second versions, which spreads the $0.125 across all of them.

A fair summary

If your catalog is thousands of sub-minute stingers and nothing else, price per minute and call the vendor directly. If you assemble ads on Sume anyway, one flat generation per mood is simple to budget. Recheck prices before you commit; both change. See Jobs and results for the job flow.