ElevenLabs has rolled out Music v2.5, which the company calls its best music generation model yet, making it the new default for both prompted and reference-based generation inside its ElevenMusic platform. The update, announced Thursday in a company blog post, pairs the quality upgrade with a set of user-facing policy changes — lossless downloads on every plan, explicit ownership language, and new restrictions on tracks that reference other artists' songs.
"Music v2.5 is now the default for prompted and reference generation in ElevenMusic," the company said. "Songs sound more layered and complex, with better, more natural-sounding instruments." For more context on this story, see our ongoing AI industry coverage.
What Changes With v2.5
The headline improvement is fidelity. According to the announcement, tracks generated with v2.5 feature richer melodies, more natural-sounding instruments, and greater depth than those produced by Music v2, which remains available for users who prefer it — v2.5 simply takes over as the default.
The second change addresses a long-standing friction point in generative music tools: getting your music out of the platform. ElevenMusic users can now download every track they finish in lossless quality, with quotas set by plan tier — five lossless downloads per day on the Free plan and 400 per month on the Pro plan. A track a user is proud of, the company's framing suggests, should not remain locked inside the tool that made it.
Ownership, Stated Explicitly
Alongside downloads, ElevenLabs has sharpened its ownership language. Users hold rights to what they create in ElevenMusic on every plan, including the free tier, and those rights attach the moment a track is created. The company states that cancelling or downgrading a subscription does not change what users own — a direct answer to the retention anxiety that follows most creative AI subscriptions.
The fine print matters, though: commercial use rights vary by subscription tier under the platform's terms. A free-tier user owns their track, but what they may legally do with it commercially is a different, tiered question. That structure mirrors conventions from stock-media and creator-tool platforms, translated to generative music.
Guardrails Around Artist References
The update also draws a new line around imitation. Downloads are blocked on tracks that reference other artists' songs — meaning users can generate material in the style of or referencing existing recordings, but cannot export those tracks out of the platform. The company frames the restriction plainly as "protecting the original creator."
The mechanism places ElevenLabs among the AI music platforms attempting to reconcile generative tools with the rights of the artists whose work colors their output. It is a narrower approach than blocking reference-based generation outright: the capability exists, but its products cannot leave the walled garden.
Separated From the Universal Music Deal
The company was careful to note that Music v2.5 is "separate" from the multi-year strategic agreement with Universal Music Group covering licensing and product development that ElevenLabs announced recently — a deal AI Buzz Wire covered earlier this week. The timing makes the distinction worth stating: as major labels negotiate licensing frameworks with AI platforms, ElevenLabs is simultaneously shipping product improvements and platform policies that operate independently of those agreements.
The two tracks — label licensing on one side, user-facing rights and guardrails on the other — together sketch the business model taking shape around AI music generation: licensed inputs at the source, monetized creation tools in the middle, and restricted outputs where generated material brushes against existing works.
A Crowded, Fast-Moving Market
ElevenLabs entered text-to-music later than some competitors but has moved quickly to position ElevenMusic as a full production environment rather than a novelty generator, layering on features such as a conversational co-writer, section-by-section song editing, and a free library of AI-generated sounds and loops in recent months. Music v2.5's combination of a quality jump with ownership and download clarity is aimed squarely at the serious-creator segment — users who treat outputs as drafts for real releases rather than disposable clips.
A Platform Building Toward Production Use
The v2.5 release is the latest step in a steady build-out of ElevenMusic as a creation environment rather than a one-prompt novelty. In late August the platform introduced Muse, a conversational co-writer, followed by Composer, a section-by-section song editing interface, and earlier in the summer a free library of AI-generated sounds and loops. Taken together, the feature set sketches a workflow — draft with Muse, refine in Composer, draw on the sound library, export lossless — that increasingly resembles the toolchain of conventional music production, with generative models substituted at each stage.
The downloads policy is the piece that makes that workflow viable outside the platform. Lossless formats preserve the full detail of a mix for mastering and distribution, and daily or monthly quotas that scale with the plan give frequent creators room to iterate without treating every experiment as a permanent export. For that audience, the meaningful news may be less the model upgrade than the package around it: lossless export, durable ownership, and a predictable rule for what cannot leave the platform. In generative music, where questions about who owns a song and whether it can be released commercially have slowed adoption among professionals, answers — even tiered, terms-of-service answers — are themselves a product feature.
Music v2.5 is live now in ElevenMusic for all users, with Music v2 remaining selectable for existing workflows.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
