Suno Debuts v6 Music Generator Built With Licensed Record Label Catalogs
The startup rebuilt its training pipeline alongside Warner Music, BMG, and Believe while introducing multimodal prompts and isolated audio editing.
Key highlights · 2 min read
- Suno has launched v6, a suite of generative audio models trained on licensed record label material rather than web-scraped archives.
- According to Jack Brody of Suno, the new architecture was "trained from the ground up, with a new set of data that does not include the same data that our previous models were trained on," speaking…
- The v6 family is split into three separate models designed for different workloads.
The Scale ReportSuno has launched v6, a suite of generative audio models trained on licensed record label material rather than web-scraped archives. The release marks the startup's first major architecture overhaul developed directly with major music industry stakeholders, following years of legal friction across the AI generation landscape.
According to Jack Brody of Suno, the new architecture was "trained from the ground up, with a new set of data that does not include the same data that our previous models were trained on," speaking to The Verge. The training dataset incorporates catalog material licensed from Warner Music Group, BMG, and Believe, alongside internal user data. The company plans to phase out its older generation models as the new lineup rolls out.
The v6 family is split into three separate models designed for different workloads. The primary v6 model handles standard high-fidelity generation, while v6-mini serves as a lightweight, faster option offered freely to all users. A third variant, dubbed v6-wild, is tuned to introduce higher variability, aiming for what Brody described as "happy accidents and natural imperfections."
Beyond training pedigree, v6 introduces conversational editing capabilities that let creators modify individual song elements, such as adjusting a specific guitar riff or revising a lyric, without forcing a complete regeneration of the audio file. Users can also blend separate generations from their Suno library together and trigger generations using images, video, and audio clips in addition to standard text prompts.
While the updated model demonstrates significantly sharper recognition of complex musical subgenres like krautrock and hyperpop, early hands-on evaluations indicate the system still struggles with human stylistic flaws. The model tends to enforce strict rhythmic and harmonic polish, frequently overriding user prompts that explicitly request out-of-tune instrumentation, discordant arrangements, or monotone vocal delivery.
Why it matters
Suno's pivot toward licensed training datasets represents a critical defensive move for generative audio startups facing mounting copyright litigation from music publishers. By formalizing data partnerships with players like Warner Music Group and BMG, Suno is attempting to build an enterprise-safe foundation while giving labels a direct stake in commercial AI music generation tools.
Reporting based on coverage from AI | The Verge.




