Music v2.5 opens commercial use to its free plan

ElevenLabs improves the instruments, voices, and arrangements of Music v2.5, while allowing the commercial exploitation of original tracks created for free with attribution.

Creating a song for free to accompany an ad, video, or branded content is now possible with ElevenLabs. With Music v2.5, the company is not simply announcing an improvement in sound quality: it is also changing the rights granted to users of its ElevenMusic platform.

Available since September 11, 2026, Music v2.5 is now the default model for creations generated from a prompt or reference track. Music v2 remains available to users who prefer its output or want to maintain continuity with earlier projects.

According to ElevenLabs, this new version produces richer melodies, more natural-sounding instruments, and arrangements with greater depth. The goal is not simply to add more elements to a track, but to make them evolve together more effectively throughout its duration.

The company particularly emphasizes productions driven by vocals and acoustic instruments. It says the clearest improvements can be heard in R&B, soul, hip-hop, rock, metal, as well as orchestral and cinematic compositions.

The word “natural” does not mean that real musicians performed the generated tracks. ElevenLabs is trying to bring their tone, variations, and dynamics closer to what might be heard in an actual recording, with fewer static sounds or layers that feel as though they were assembled separately.

The company compared Music v2 and v2.5 in a blind test involving 47,885 pairs. The same prompt was used to produce one version with each model, and participants were then asked which one they preferred. Music v2.5 reportedly won the majority of comparisons.

That result offers an initial indication, but remains incomplete. ElevenLabs does not disclose the exact preference rate, the number or profile of participants, or the prompts used. No test set or detailed protocol has been published that would allow the evaluation to be reproduced independently.

The claimed improvement builds largely on features introduced with Music v2 in May 2026. The model can construct a track section by section, maintain longer compositions, switch genres mid-song, and handle fast or highly dense lyrics, particularly in rap.

Generation can begin from a mood, style, tempo, instrumentation, or lyrics. Users can request a vocal or instrumental track, define its structure, and specify how its different sections should evolve. Creations can range from three seconds to five minutes, according to the Music documentation.

Genre transitions can, for example, begin with an orchestral introduction, move into a rock section, and then shift into electronic production. The model does not necessarily place three independent clips next to one another: it attempts to preserve musical continuity throughout the transitions.

That ability still depends on the precision of the request and the complexity of the track. A convincing transition requires the rhythm, timbres, key, vocals, and structure to evolve simultaneously. ElevenLabs has not yet published a dedicated benchmark comparing the reliability of these transformations between v2 and v2.5.

The handling of multilingual lyrics and vocals is another area being highlighted. The model is intended to produce voices that sound as though they belong to the requested language, with more consistent pronunciation and phrasing. The announcement does not, however, provide a complete list of the languages evaluated or detailed results for each one.

ElevenLabs’ wording that voices should sound “native” in the user’s language should therefore be understood as a quality claim rather than a guarantee. It does not ensure the absence of unintended accents, pronunciation errors, or distorted syllables, particularly with fast lyrics, proper names, or less represented languages.

Music v2.5 also retains inpainting. This feature allows users to select part of an already generated track and regenerate it without starting over. A verse can be changed, a bridge replaced, or a transition reworked while preserving the rest of the composition.

Continuity around the replaced section remains an important challenge. The new passage must match the tempo, key, acoustics, and vocal characteristics already established. ElevenLabs says v2.5 improves overall quality and prompt adherence, but has not published a dedicated evaluation of the transitions produced by this feature.

Another mode uses an audio excerpt as a reference. It accepts around 30 seconds of audio to guide the production, instrumentation, tempo, mood, and overall character of a new song. This feature is available with both v2 and v2.5 on paid plans.

The uploaded track should not be understood as material that is directly incorporated into the new recording. According to the documentation, Audio Reference guides the creation without copying or remixing the recording. Each file is checked for protected content.

This distinction makes the system less suitable for radical transformations. Uploading a jazz song and then asking for a rap version does not match the intended use case and may produce less reliable results. The reference is primarily designed to indicate a sonic direction close to the supplied excerpt.

The messaging published on X uses broader language, suggesting that users can start from an idea, a sound, or a loop and turn it into a finished song. In practice, the outcome depends on the origin of the file. An excerpt created by the user or one for which they hold the rights can be used as a reference, while a commercial track may be rejected or lead to export restrictions.

Music v2.5 is being deployed across three environments. ElevenMusic is aimed primarily at musicians and general users. ElevenCreative allows tracks to be integrated into ads, videos, and branded content. Music API provides the same capabilities to developers who want to automate generation within their own products.

Within ElevenCreative, the model can be used in Studio to create a soundtrack for a video. It can also become a step inside Flows, the environment where multiple creative tools are chained together. A team