ElevenLabs Music 3 s to 5 min vs Sume Music: how to set length

ElevenLabs Music accepts a length between 3 seconds and 5 minutes, while Sume Music has no length field at all: you describe the length in the prompt, and a duration or duration_seconds field is rejected. If you need an exact 3-second sting from Sume, generate a short brief and cut it afterwards.

The ElevenLabs numbers come from its Music documentation (read 2026-10-10). Sume's come from the Music 1.0 and Music Router pages.

What each side lets you set

ElevenLabs lists Music v2.5 as the default, with v2 and v1 still available during a transition, a 3-second minimum and a 5-minute maximum, and MP3 (44.1 kHz, 128 to 192 kbps) or WAV output. Sume's Music Router runs Lyria 3.5 behind sume/music-auto; the docs describe full-length structured songs of up to a few minutes, steered by the prompt.

Setting length on Sume without a field

The Music 1.0 page says to write the length into the prompt, for example 'a 2-minute track', or to use section markers with timestamps. It also says the model-reported lyrics or section map in the result are metadata, not an audio measurement, so check the actual file duration before you build a timeline around it.

Because the price is fixed per generation, a 20-second brief costs the same $0.125 as a long one. That favours generating a slightly longer track than you need and trimming it.

Why the missing field is a design choice

A duration field implies the output will land on that length. Sume's Music page takes the opposite stance: it tells you the length is steered by the prompt and the result should be examined. A field that rejects duration with an error is clearer than one that silently ignores it, because your code finds out on the first call instead of after a paid generation.

If your own tooling already sends a duration to other music vendors, map it into the prompt text on the Sume side: 'a 45-second track' for the brief, plus the exact-length trim step afterwards. Keep the requested length in your job metadata so you can compare it with the file you actually got. The metadata field is stored on the Sume job and is not sent to the provider, which makes it a safe place for that note.

Getting an exact length

For an exact cut, use the audio tools. Timeline audio can split a Sume-hosted file into ranges: up to 20 ranges per job, one flat $0.01 per job, with WAV output as the sample-exact default (Timeline audio). A 3-second sting from a 30-second generation is one split range, { "start": 0, "end": 3 }.

Import the Music artifact first if the split job rejects the URL as off-host; timeline audio only accepts workspace media.sume.com audio.

The total for a trimmed sting

One generation at $0.125 plus one split job at $0.01 is $0.135 for an exact 3-second sting. The same arithmetic for a 5-minute ElevenLabs track depends on your plan's credits, which that page does not state, so this post does not compare dollars. For per-minute dollar comparisons see cost per finished minute.

• Need under 3 seconds with a length field: ElevenLabs lists 3 seconds as its floor, so it cannot do that either.
• Need a precise loop: generate a longer track, split the best bar boundary, and loop it in a timeline render.
• Need over 5 minutes: join tracks; see longest AI song length.