Stable Audio 3.0 Creates Six-Minute Songs

Stability AI has released Stable Audio 3.0, a new family of generative audio models that can create songs as long as six minutes and 20 seconds, moving AI music generation closer to full-length commercial composition rather than short experimental clips. The company says the models are trained on fully licensed data, a positioning that directly addresses one of the biggest legal and business concerns hanging over the AI music industry.

The release comes as AI-generated music shifts from novelty demos to a more serious creative-production market. Suno, Udio, Google, Nvidia, ElevenLabs, Adobe, and other AI audio players are all competing to give creators faster ways to generate songs, sound effects, stems, and background scores. Stability AI’s latest move stands out because it combines longer generation, open-weight availability for several models, and a licensing narrative designed to appeal to developers and enterprises.

Longer Tracks, Open Weights and Local Generation

Stable Audio 3.0 includes four models: Small SFX, Small, Medium, and Large. Stability AI says Small SFX is aimed at sound effects, Small can generate full music on-device, Medium supports longer and more musical tracks, and Large is designed for high-volume creative platforms and enterprise use. The Small SFX, Small, and Medium versions are available as open-weight models, while Large is available through Stability AI’s API and enterprise self-hosting.

The headline improvement is duration. Stable Audio 3.0 Medium and Large can generate tracks up to six minutes and 20 seconds, while Small can generate up to two minutes. That is a major jump from Stable Audio Open Small’s 11-second limit and Stable Audio Open’s 47-second limit, according to Stability AI. The company also says the new architecture uses a semantic-acoustic autoencoder to support longer, more flexible audio generation with second-level control.

Licensing Becomes a Competitive Feature

Stability AI is leaning heavily into the claim that Stable Audio 3.0 was trained on fully licensed data. That matters because AI music tools are operating in a legally sensitive market where artists, labels, and publishers are increasingly focused on training data, output ownership, and commercial usage rights. Stability AI says users own their outputs under its Community License, while companies with more than $1 million in annual revenue need an Enterprise License.

That licensing approach gives Stability AI a clearer enterprise pitch. For brands, agencies, game studios, video creators, and music platforms, the question is not only whether a model can produce a good track; it is whether the output can be used without triggering copyright, attribution, or indemnity concerns. Stability AI says its Enterprise License includes legal indemnification, which could make Stable Audio 3.0 more attractive to commercial teams that cannot rely on unclear training-data provenance.

Competition Moves Beyond Simple Text-to-Music

The AI audio market is getting crowded quickly. Suno and Udio have become closely associated with consumer-facing AI song generation, while Nvidia’s Fugatto has been positioned as a research-heavy model that can generate music, transform voices, and create unusual sound effects, though Nvidia has not publicly released it. Google, Adobe, ElevenLabs, OpenAI, and others are also exploring different parts of the audio-generation stack.

Stable Audio 3.0’s opening is not just song length. Its strategic value is in giving developers and creative companies models they can download, fine-tune, and integrate into workflows. Stability AI is also releasing LoRA training documentation for Small and Medium, which could help users adapt the models to specific sounds, genres, or production needs.

Broader Impact on Creative Production

The larger impact is that AI music is moving toward production infrastructure. A six-minute generation window makes the technology more useful for podcast beds, YouTube background tracks, game loops, advertising music, meditation audio, product videos, and creator soundtracks. It also changes the economics of sound production by making original audio cheaper and faster to generate at scale.

The trade-off is that music quality, originality, artist compensation, and disclosure will remain difficult questions. Even with licensed data, AI-generated music challenges traditional workflows for composers, stock-audio libraries, session musicians, and production houses. Stability AI is trying to frame Stable Audio 3.0 as a tool for experimentation and commercial creation, but adoption will depend on whether creators see it as a useful assistant or another technology compressing the value of human-made music.

For now, Stable Audio 3.0 gives Stability AI a stronger place in the AI music race. The company is no longer only competing on generation quality; it is competing on track length, open access, local execution, licensing confidence, and enterprise readiness. That combination may matter more than a single viral demo as AI audio becomes part of everyday creative software.