ElevenLabs Music: The Licensed-Training AI Music Model, v1 to v2 Explained (August 2026)
Last updated August 7, 2026
ElevenLabs Music generates full songs (10 seconds to 10 minutes, vocals or instrumental) and is the licensing-first AI music model: trained on licensed data via Merlin and Kobalt deals, every track cleared for commercial use — though self-serve plans exclude film, TV, and Studio Games; Enterprise covers all. Music v2 (May 2026) added genre transitions and section inpainting. V1 and V2 run in the invideo agent's Music tab.
Updated August 2026
ElevenLabs Music generates complete, studio-grade songs — vocals or instrumental — from a text prompt, at any length from 10 seconds to 10 minutes. Its defining trait is not a spec but a legal position: ElevenLabs states the model is "trained on licensed data only" and that "every track you generate is cleared for commercial use," backed by licensing deals with Merlin Network and Kobalt Music Group. One nuance matters enormously and is widely misquoted: on self-serve plans the commercial clearance excludes film, TV, and Studio Games — Enterprise plans cover all commercial use. Both Music v1 and v2 run inside the invideo agent's Music tab.
What is ElevenLabs Music and how did it get here?
Three dated milestones, all from ElevenLabs' own announcements:
- August 5, 2025 — Eleven Music (v1) launches, positioned from day one on cleared rights rather than raw audio quality.
- April 29, 2026 — ElevenMusic arrives: a consumer discovery, remix, and monetization layer built with more than 4,000 independent artists.
- May 26, 2026 — Music v2 ships with better vocals and arrangement, genre-shifting within a single track, fast and dense rap delivery, embedded sound effects, section inpainting, improved multilingual reliability — and API price cuts of up to 50%.
Most users want v2; v1 remains available as music_v1.
Why is the ElevenLabs licensing story different?
Most AI music vendors either fought the labels or stayed silent about training data. ElevenLabs negotiated instead, announcing launch-day deals with Merlin Network, the independent-label licensing body, and Kobalt Music Group. Merlin CEO Jeremy Sirota framed the deal directly: "Our partnership demonstrates that music rightsholders can negotiate thoughtful, forward-looking agreements with AI companies," per Music Business Worldwide's launch coverage. At a time when other music models face active label litigation, that makes this the cleanest provenance story in AI music as of August 2026.
Now the part to get exactly right — launch-era language and current terms differ. The elevenlabs.io/music page states that self-serve commercial use covers online and offline commercial use "except for film, TV, and Studio Games"; Enterprise plans permit all commercial use. So a marketing video, a YouTube channel, a podcast, an ad — cleared on a paid self-serve plan. A film score, a broadcast TV placement, or a AAA game soundtrack — that requires Enterprise. The August 2025 launch post's broader "nearly all commercial uses" phrasing is superseded by the current carve-out.
Guardrails match the licensing posture: prompts cannot reference artist or stage names, song, album, or label names, or reproduce substantial existing lyrics. Outputs carry C2PA content-provenance signing.
Which ElevenLabs Music variant should you pick?
invideo's Music tab exposes the family as three picker entries:
| invideo picker entry | Model | Output | Best for |
|---|---|---|---|
| ElevenLabs Music V2 Song | music_v2 |
Full song with vocals and lyrics | Theme songs, lyric-driven tracks, rap, multilingual vocals |
| ElevenLabs Music V2 Instrumental | music_v2 with force_instrumental |
Guaranteed no-vocals track | Background beds, scores, anything under a voiceover |
| ElevenLabs Music V1 | music_v1 |
Song or instrumental | Continuity with earlier v1 outputs |
The Song/Instrumental split is not two models — it is one v2 model with the force_instrumental flag set, per the official compose API reference.
What are ElevenLabs Music's specs?
| Spec | Value (Aug 2026, official API docs) |
|---|---|
| Track length | 3,000–600,000 ms (10 seconds to 10 minutes) |
| Models | music_v1, music_v2 |
| Vocal control | force_instrumental flag; otherwise vocals with lyrics |
| Composition | Simple prompt or structured composition_plan (section-by-section; mutually exclusive with prompt) |
| Editing | Section inpainting (store_for_inpainting); v2 adds genre transitions within a track |
| Output formats | MP3 (32–320 kbps), PCM, Opus, µ-law/A-law, 8–48 kHz |
| Provenance | C2PA signing; seed control for reproducibility |
The row that matters most is track length: 600,000 ms means a single generation can outlast most of the videos it will score, so you compose to the edit rather than looping a 30-second cue.
Two further capabilities round out the toolkit as of August 2026: steering generations with a short audio reference (around 30 seconds) so a track matches an existing sonic direction, and Music Finetunes for training a custom style on roughly 5–10 minutes of your own audio. Both are newer additions — check current ElevenLabs docs for exact limits.
What does ElevenLabs Music cost, and when is commercial use included?
From the official pricing page: the Free tier ($0) does not include music commercial use; commercial rights start at the Starter tier ($6/month) and continue through Creator ($22), Pro ($99), Scale ($299), and Business ($990), with credits shared across all ElevenLabs products. Remember the scope: these self-serve tiers exclude film, TV, and Studio Games — those need an Enterprise agreement. Music v2's launch also cut API prices by up to 50%, per the v2 announcement.
What are the documented strengths and limits?
Strengths. The licensing chain is the moat: licensed training data, named rightsholder deals, C2PA-signed outputs, and commercial clearance from a $6 tier. Ten-minute generations are the longest among the major music models in invideo's roster, and composition plans plus section inpainting give real structural control — you can rewrite one verse without regenerating the song.
Limits. The film/TV/Studio Games carve-out on self-serve plans is the one that bites professionals, and it is easy to miss. Music credits draw from the same pool as ElevenLabs' voice products, so heavy music use competes with voiceover budgets. And ElevenLabs publishes no head-to-head audio-quality benchmarks, so quality claims rest on demos rather than dated third-party numbers.
What can you make with ElevenLabs Music?
The natural fit is any video where the track carries the piece: generate a full-length song and cut a music video against it, score a brand film with a v2 instrumental bed, or build ad variants by inpainting a new chorus into a proven track. Because commercial clearance starts so low, small-channel YouTube and podcast intros are a sweet spot. If you are here for ElevenLabs' voices rather than its music, that is a separate model family with its own page — see the ElevenLabs voice models guide for TTS, voice cloning, and dubbing.
How do you prompt ElevenLabs Music?
For quick results, a single descriptive prompt works: name the genre, mood, instrumentation, tempo, and whether you want vocals ("melancholy acoustic folk, female vocal, fingerpicked guitar, slow, about leaving home"). For control, skip the prompt and supply a composition plan — section-by-section structure specifying each part's style and lyrics; the two approaches are mutually exclusive. Write lyrics into the plan when the words matter; use force_instrumental when they don't. If a section misses, inpaint that section rather than rerolling the track.
The fine print: ElevenLabs Music questions
Can I use ElevenLabs Music commercially? Yes, from the Starter tier ($6/month) upward — for online and offline commercial use except film, TV, and Studio Games, which require an Enterprise plan. Free-tier tracks are not cleared for commercial use.
Is ElevenLabs Music trained on licensed data? Yes. ElevenLabs states it is "trained on licensed data only," with deals covering Merlin Network's independent-label catalog and Kobalt Music Group's publishing.
How long can an ElevenLabs song be? 10 seconds to 10 minutes (3,000–600,000 ms), per the official API reference.
Can it make instrumental-only tracks? Yes — the force_instrumental parameter guarantees no vocals; invideo surfaces this as the ElevenLabs Music V2 Instrumental picker entry.
What's new in Music v2 vs v1? Better vocals and arrangement, genre transitions within one track, fast/dense rap delivery, embedded sound effects, section inpainting, stronger multilingual output, and API prices cut up to 50% (May 26, 2026 announcement).
Can I edit part of a generated song? Yes — section inpainting lets you regenerate a chosen section while keeping the rest, if the track was stored for inpainting.
Where can you use ElevenLabs Music?
Inside invideo, ElevenLabs Music sits in the agent's Music tab as three entries — ElevenLabs Music V2 Song, ElevenLabs Music V2 Instrumental, and ElevenLabs Music V1 — so the same agent that sequences your shots across 200+ models can also score them with a commercially cleared track, no separate subscription required. The ElevenLabs hub covers the family's full presence on the platform, and the music video maker is the shortest path from generated song to published video.
Version history: first published August 2026, covering Eleven Music v1 (Aug 2025), the ElevenMusic platform (Apr 2026), and Music v2 (May 2026). Will be updated if the self-serve commercial-use scope or pricing tiers change.