Chapter 27 · Eleven Music

Yasser Ashour

2 hours ago

8 min read

Part 5 · Music and sound

Suno is the first route to a cue. Eleven Music, on the ElevenLabs account the voice work already uses, is the alternative when a cue must hit timed sections and the delivery is online. You write it as a plan of chunks and repair it by keeping a stored range exactly. Its self-serve licence excludes film, television and radio.

In this chapter

What Eleven Music 2.5 is, and when to choose it over Suno
Prompt mode and plan mode; a cue written as chunks and repaired by keeping a stored range
The settings, the checks, and the limits, prices and licence

Before you start. Chapter 25 decides the cue and its windows; Chapter 26 is the first route. The same brief is translated here into chunks.

27.1 What it is

Eleven Music is ElevenLabs' music model, reached in the ElevenCreative app and through the ElevenLabs API. Its current model is Music 2.5, which ElevenLabs calls "our most advanced music model" (documentation checked 30 Sep 2026). The API lists it as music_v2_5 (changelog of 14 Sep 2026); music_v2 and music_v1 remain available. The API still defaults to music_v1 and the pages disagree about the app's default, so name the model every time.

Choose it when a cue needs sections of known length, such as a 30 second online spot with a change at second 24, and the delivery is online. Choose something else when the delivery is film, television or radio (a composer, a licensed library, or Suno on a paid plan, Chapter 26), or when you want a single sound event (Chapter 28).

27.2 How it reads what you give it

A request uses one of two modes, never both (checked 30 Sep 2026).

Prompt: a whole cue from one description, up to 4,100 characters, with music_length_ms from 3,000 to 600,000 and force_instrumental.
Composition plan: an ordered list of up to 30 chunks. Each has text, duration_ms (3,000 to 120,000), positive and negative styles (up to 50 each) and a context-adherence level.
Audio reference chunk: a stored range of a song, kept exactly; the song must have been stored for inpainting.
Conditioning reference: a chunk regenerated "staying close to the reference's musical characteristics"; up to 30 seconds, strength low, medium, high or xhigh.
Video to music: up to ten video files (200 MB and 600 seconds in all), a description up to 1,000 characters and up to ten tags; no field for cut points.

The maker's prompting page says the model reads five dimensions: genre, mood, instrumentation, tempo and production era. "Any question you leave open, the model answers with the most statistically likely choice — which is to say, the most average one." Chunks bind by order. The first chunk's styles "are the most important as they set the overall tone and genre"; context adherence says how closely a later chunk follows its neighbours.

27.3 The prompt, in this tool's dialect

Write the brief of 25.3 as roles and behaviour, then split it into chunks measured from the picture. Its eight parts all apply; the theme goes in the first chunk.

A chunk's text holds a section name in square brackets and inline directions in braces: [Intro] {claps alone, then the riff}. Character, instruments and feel go in the style lists.
"Instrumental" and the vocal refusals go in every chunk: force_instrumental belongs to prompt mode only.
Never name an artist, a band or a lyric. Copyrighted material in a style returns an error (bad_composition_plan for a plan, bad_prompt for a prompt).
Nothing shorter than three seconds, and a hit is not a field. Make a one-second sting in the edit, and put a must-hit click on the effects track (25.10).
Keep and remake are different requests. To keep approved music, put its stored range into the plan as an audio-reference chunk. To write new music that resembles it, give a new chunk a conditioning reference.

27.4 Templates

Written for this book · template; not run · finished grade, a cue as a plan (human-readable; compile it into the request's fields)

Template — the plan

MODEL: music_v2_5 · composition plan · store for inpainting: true · output format: [named]
Chunk n: [ms]
  text: [Section] {[short direction]}
  positive_styles: instrumental, [ensemble, feel, mode, tempo]
  negative_styles: vocals, [what must not appear]
  context_adherence: [high / medium / low]

Written for this book · template; not run · test grade, a prompt for a temp cue

Template — the test prompt

[Genre], [mood], [instrumentation], [tempo], [production era]. Instrumental only. [How it starts], [how it ends].
force_instrumental: true · music_length_ms: [n] · model_id: music_v2_5

27.5 Examples at the standard

Written for this book · telecom and tech · a 20-second vertical online promo for a data top-up, Cairo, voice-over in the middle chunk · Eleven Music 2.5 (`music_v2_5`), composition plan, store for inpainting on · written for this book; not run

Example 27.1 — a five-note riff as a timed plan

MODEL: music_v2_5 · composition plan · store for inpainting: true · output format: mp3_44100_192
Chunk 1: 4000 ms
  text: [Intro] {claps alone, then a five-note oud riff stated once}
  positive_styles: instrumental, Gulf hand-clap groove, swinging six-beat pulse, close dry oud, crisp claps, G minor with a raised seventh, about 104 BPM
  negative_styles: vocals, choir, trailer risers, EDM drop, orchestral strings
  context_adherence: high
Chunk 2: 12000 ms
  text: [Under the voice] {riff thinned to two repeated notes, claps continue, middle register empty}
  positive_styles: instrumental, same ensemble, sparse middle register, round sub-bass enters softly, steady 104 BPM
  negative_styles: vocals, busy lead melody, riser, choir
  context_adherence: high
Chunk 3: 4000 ms
  text: [Ending] {full riff with one plucked synth, hand drum joins, plain cadence on G, last clap and a half-second tail}
  positive_styles: instrumental, same riff in full, plain cadence, clean stop
  negative_styles: vocals, fade, long reverb tail, drum fill
  context_adherence: medium

The first chunk carries the palette: ensemble, mode and tempo are written once, so later chunks say "same ensemble".
Each chunk is a section with a job. The voice-over gets its own chunk; lengths come from the picture map.

Written for this book · telecom and tech · the smallest repair of Example 27.1: keep the first 16 seconds, remake a too-busy ending · Eleven Music 2.5, audio-reference chunk plus one new chunk · written for this book; not run

Example 27.2 — keep what is approved, remake the last four seconds

MODEL: music_v2_5 · composition plan
Chunk 1: audio reference · song_id: [id of the stored song] · range: 0 to 16000 ms, kept exactly
Chunk 2: 4000 ms
  text: [Ending] {the riff once in full, one plucked synth doubling it, then one clap and stop}
  positive_styles: instrumental, same ensemble, same riff, plain cadence on G, dry and tight
  negative_styles: vocals, drum fill, riser, reverb tail, fade
  context_adherence: high

"Keep this" is an audio-reference chunk, not a prompt sentence; only the last four seconds are generated.
Measure the boundary first, at a phrase edge on the returned file, or the seam clicks.

27.6 Settings

Setting	Value to use (checked 30 Sep 2026)	Why
model_id	music_v2_5, named in every request	the API default is music_v1, which writes sections, not chunks
Mode	a plan for timed sections; a prompt for a temp cue	one per request
store_for_inpainting	true, at generation	default false; without it an ending cannot be kept and repaired later. If you forgot, upload the approved file to store it
output_format	name it; default auto	MP3 at 192 kbps needs the Creator plan or above, PCM at 44.1 kHz the Pro plan or above
force_instrumental, music_length_ms	prompt mode only	in a plan, the chunks carry both
seed	optional; not with a prompt	it "can help" repeat a result; no guarantee

27.7 Checks before you sign

Before an Eleven Music cue goes near a client

The delivery is online, or an Enterprise Music plan is in hand and ElevenLabs has confirmed in writing that its terms cover Music 2.5 (27.9).
The plan's eligibility row fits whoever holds it.
Listen alone on headphones: no voice, no palette leak into the ending, the figure where the chunk says, and clean chunk joints and repair seams.
Measure where each phrase and the ending land. Probe the file: format, sample rate, length. Loudness is measured in the mix (25.14).
The record holds the plan, the song ID and the rights line (34.10).

27.8 Failures and fixes

You hear or see	Cause	Smallest fix
It sounds older, or ignores the chunks	it ran on the default music_v1	name music_v2_5
A voice appears	force_instrumental does not apply to plans	"vocals" in the negative styles of every chunk
The ending cannot be repaired	it was not stored at generation	upload the approved file to store it, then repair
Chunk 1's palette leaks into a changed ending	the first chunk shapes all that follow	an audio-reference chunk for what must stay; lower context adherence on the chunk that should change
A click at a chunk joint, or a refusal	the boundary sits mid-phrase; or a name in a style	move the reference range to a phrase edge; describe the sound instead of naming it

Retry ladder. Measure the fault by chunk; change one style phrase; lower context adherence on that chunk; repair with an audio-reference chunk; then Suno or a composer.

27.9 Limits, prices and rights

Limits (checked 30 Sep 2026). A plan holds up to 30 chunks of 3 to 120 seconds each and runs from 3 seconds to 10 minutes.

Prices. Music uses the account's credits; hover over the remaining-credits display before generating to see the cost. The pricing FAQ (30 Sep 2026) puts Eleven Music at approximately 900 credits a minute; the API bills $0.15 a minute. The pricing page shows Free at $0 with 10,000 credits a month, Starter $6 with 30,000, Creator $22 (first month $11) with 121,000, Pro $99 with 600,000, and Scale and Business above that. The terms cap generation each month at 11 minutes on Free, 17 on Starter, 62 on Creator and 304 on Pro, and downloads at none, 30, 250 and 500.

Rights: the model-specific terms (updated 26 May 2026, read 30 Sep 2026). They apply to the music_v1 and music_v2 families, subversions included, unless ElevenLabs designates a subversion a new Version. Music 2.5 is not named; its id, music_v2_5, puts it in the music_v2 family. Every self-serve plan and Enterprise Music Lite reads: "All online and offline commercial use permitted, except film, TV, radio, & Studio Games". Only Enterprise Music reads "All online and offline commercial use permitted".

Eligibility: Free, Starter, Creator and Pro are "For Individual Use Only"; Scale is for individuals or organisations with fewer than 10 employees, Business fewer than 50; the Enterprise plans have no limit. Free requires attribution and permits no downloads, and the pricing page lists music commercial use from Starter, so Free is a trial, not a client route. Free and Starter prohibit streaming; reseller arrangements and music libraries for licensing are barred. The Music Terms bar customers in weapons, tobacco, prescription drugs, adult entertainment, religious organisations and political campaigning. Before any final beyond online, ask ElevenLabs in writing whether Music 2.5 is a new Version. The product page calls the music "cleared for nearly all commercial uses, from film and television"; the terms win. Tell the client in writing that a generated cue is not an exclusive score (34.9).

27.10 Version notes

What Music 2.5 changed (September 2026). ElevenLabs says its songs have "more layers, more movement" and "hold together all the way through". music_v2_5 is accepted on the planning, composition, inpainting, video-to-music, upload and fine-tuning endpoints (changelog of 14 Sep 2026). The API default is still music_v1; ElevenLabs says Music 2 will become the default after a transition period.

Re-check when the next model ships: the model IDs and API default; whether the terms name the new version; the Music pricing row; the maximum track length.

Open questions that change what you do

Whether ElevenLabs has designated Music 2.5 a new Version under the terms of 26 May 2026. Ask in writing before any use beyond online.
How long a track may be. The overview says 5 minutes, the API 10. Test before planning a long cue.
Whether it sings Egyptian or Saudi dialect correctly. Not established; sung Arabic goes to a reviewer.

What to remember

Use Eleven Music when a cue needs timed sections and the delivery is online; Suno is the first route.
Name music_v2_5 in every request. Send a plan or a prompt, never both.
Write the brief as chunks: the first carries the palette, each has a job, and "instrumental" and the vocal refusals go in every chunk.
Store for inpainting at generation. Keep approved music with an audio-reference chunk; remake with a conditioning reference.
Measure the boundaries and the hits; there is no hit-point field.
Self-serve plans exclude film, TV and radio; only Enterprise Music lifts that.