ElevenLabs Music v2 Can Switch Genres Mid-Track, Built on Licensed Data

ElevenLabs launched Music v2 on Wednesday, a generative AI music model that can shift genres inside a single track — going from opera to heavy metal and back without losing musical coherence. The model is available now in ElevenLabs' ElevenCreative tool and the new ElevenMusic platform, with API access coming soon.

The headline capability sounds like a parlor trick. It isn't. Mid-track genre switching, fast rap that stays legible, structured section-by-section composition — these are the technical features that move generative music from "fun demo" to something a working musician or marketing team can actually use in a project.

The Context: Generative Music Is Finally Catching Up to Image and Video

For most of the AI boom, music has been the laggard category. Image generation crossed the "good enough for production" threshold in 2023. Video crossed it in 2025 with OpenAI's Sora 2 and Google's Veo 3. Music kept producing demo-quality output — fine for a 15-second TikTok bed but not for anything a professional would ship.

The bottleneck wasn't raw model capability. It was structure. Music requires long-range coherence — the same melodic theme has to develop across a three-minute song, the tempo has to be consistent, the transitions between sections have to feel intentional. Earlier models generated short clips well but fell apart on longer compositions.

Music v2 is the first model from a major lab that explicitly addresses this. ElevenLabs trained the system to compose by sections — intro, verse, chorus — that stitch together into a coherent whole. Artists can also pick a single part of an existing track and regenerate it with new prompts without touching the rest. That's a workflow primitive that didn't exist in generative music before.

The Details: What Music v2 Actually Does

The new capabilities ship as a tight bundle. The model handles complex vocal performance — opera, screaming metal vocals, fast multi-syllable rap — and can switch between them mid-song without artifacts. It can layer non-musical sound effects into a track (a door slam at a beat drop, ambient noise during a quiet section) in a way that sounds like a producer placed them, not a model bolted them on.

Section-by-section composition is the bigger deal for working creators. Instead of generating a full track and hoping the model nails every part, you can specify the intro one way, the verse another, the chorus a third, and stitch them. Each section can be regenerated individually. That mirrors how real producers actually build songs — and it solves the "the song is 80 percent right but the bridge is wrong" problem that's killed most music AI workflows for paying customers.

The third capability is targeted re-creation. Pick a slice of an existing track — a guitar solo, a chorus, a transition — and prompt the model to rebuild just that slice in a different style without touching the surrounding audio. For sync licensing, sound design, and remix work, that's the killer feature.

Industry Impact: The Licensing Story Matters More Than the Tech

Music v2's most consequential detail isn't the model architecture. It's the data. ElevenLabs says the model was trained on fully licensed material, and outputs are cleared for commercial use. Customers can use generated tracks in ads, films, games, and products without copyright exposure.

That puts ElevenLabs in a different category from competitors like Suno and Udio, both of which are mid-lawsuit with major labels over training data. RIAA member labels sued both companies in 2024 over unlicensed training, and those cases remain unresolved. Enterprise customers — ad agencies, gaming studios, film production companies — have largely sat out generative music because of the legal uncertainty.

Music v2 is the first model from a credible vendor that explicitly removes that risk. Why this matters: it opens the entire enterprise creative budget to AI music. A Fortune 500 marketing team can now use generated music in a global campaign without a legal review eating the timeline.

Suno and Udio remain the leaders on raw model quality and viral consumer adoption, but they're locked out of the highest-value commercial use cases until the lawsuits resolve. ElevenLabs has positioned itself to absorb that enterprise demand right now.

Expert Perspectives

Working musicians and producers reacted to Music v2 with the same mix the industry has had for the last two years: appreciation for the technical leap and anxiety about the labor implications. The genre-switching capability in particular hits a niche — sync licensing for film and TV — that previously required hiring a composer.

Audio engineers noted that the section-stitching workflow is the actual breakthrough. Generative music models that could only emit full tracks were impractical for production work because real songs go through dozens of revisions. A model that lets you regenerate a four-bar section without touching anything else is the first generative audio tool that fits inside a normal DAW workflow.

What's Next

ElevenCreative subscribers get Music v2 immediately. The standalone ElevenMusic platform — aimed at solo creators and indie artists — is also live now. API access is coming, which is the access tier that matters most for developers building music features into other products.

Expect Suno and Udio to respond within weeks. Both companies have been working on long-form coherence and licensed-data versions of their models. The interesting question is whether either can get to commercial-cleared output before their lawsuits force a settlement that resets the entire training-data question for generative music.

Watch also for major labels' response. Warner, Universal, and Sony have been negotiating licensing deals with multiple AI music vendors. A successful ElevenLabs launch could accelerate those negotiations.

Bottom Line

ElevenLabs just shipped the first generative music model that's good enough for professional use and legally clean enough for enterprise use. Mid-track genre switching is the demo that gets attention. Licensed training data is what actually changes the business. If you run a creative team, this is the model that lets you start using AI music in client work without a lawyer in the room.