ElevenLabs Music length: 3 seconds to 5 or 10 minutes vs Sume
ElevenLabs' two pages disagree on max Music length: 5 minutes on the capabilities page, 600000 ms on the Compose reference. Sume has no length field at all.

ElevenLabs Music has a minimum length of 3 seconds. Its capabilities page gives a 5 minute maximum, while its Compose API reference allows music_length_ms up to 600000, which is 10 minutes. Sume's Music Router has no length field: you steer length in the prompt and trim in a render.
Two ElevenLabs pages, two maximums
The capabilities page says generated music has a minimum of 3 seconds and a maximum of 5 minutes (read 2026-10-02). The Compose reference gives music_length_ms a minimum of 3000 and maximum of 600000, used only together with prompt (read 2026-10-02). If you need the longer figure, test it before relying on it.
| Page | Minimum | Maximum |
|---|---|---|
| Music capabilities | 3 seconds | 5 minutes |
| Compose API reference | 3000 ms | 600000 ms |
| Sume Music Router | No field | No field; steer in prompt |
What Sume does
Music Router rejects duration and duration_seconds. The docs tell you to write length into the prompt, such as a 2-minute track, or to use section markers like [0:00-0:30] Intro:. Music 1.0 on Lyria 3.5 produces full songs up to a few minutes.
Your first paragraph of the brief can state the length, and the last can say Instrumental, no vocals. The docs treat these as creative directions, so measure the file you get.
Getting an exact length anyway
Put the track into a Timeline 1.0 render. audio.duration_seconds sets the output length from 1 to 1800 seconds, and the soundtrack field takes loop and fade_out_seconds up to 10, so a bed shorter than the video can repeat and a longer one is cut by the render and faded out.
That is how you hit a 15 or 30 second slot without a duration parameter. You pay for the Music generation once, and the render handles the cut.
Sources
Related posts
More in Comparisons
- ElevenLabs TTS output_format list vs Sume container and encoding
ElevenLabs names output_format strings like opus_48000_64 and ulaw_8000. Sume splits it into container, sample rate, bit rate and encoding, with no Opus.
- ElevenLabs Professional Voice Clone: own voice only, 24 h retry
ElevenLabs Professional Voice Cloning only accepts your own voice, needs 30 minutes of audio and a verification step. What Sume's Voices library asks instead.
- ElevenLabs use_pvc_as_ivc: what it does and Sume's voice selector
ElevenLabs' use_pvc_as_ivc flag swaps a professional voice clone for its instant version. Sume's TTS has no such switch: you pick an avatar or a voice id.
- ElevenLabs stability 0.5 and similarity 0.75 vs Sume generation_config
ElevenLabs defaults stability to 0.5 and similarity to 0.75. Sume has no such sliders: generation_config takes only volume, speed and an emotion string.
Written by Sume