Stability AI releases Stability Audio 3.0 with 6-minute song generation

At a glance:

• Stability AI launches Stability Audio 3.0, enabling generation of professional-grade songs up to six minutes long.
• Four models: small SFX (459M params), small (459M params), medium (1.4B params), and large (2.7B params); small models for on-device use, medium and large for full compositions.
• Open weights for small models; large model via API with enterprise licensing required for companies earning over $1 million annually.

Stability AI unveils Stability Audio 3.0

Stability AI, the company behind the popular Stable Diffusion image model, is expanding into audio with the release of Stability Audio 3.0. This new family of models promises significant advancements in AI-generated music, with the top model capable of producing songs longer than six minutes. The company claims that these compositions can maintain musical structure and melodic tone, marking a leap from previous versions.

The announcement comes as the AI music space heats up, with competitors like Google and ElevenLabs also releasing related tools. Stability AI's move into longer-form audio generation could set a new standard for what synthetic media can achieve in creative industries.

Model specifications and capabilities

The Stability Audio 3.0 suite includes four distinct models tailored for different use cases. The small SFX model, with 459 million parameters, is designed for generating sound effects and short audio clips. The small model, also 459M parameters, suits on-device music generation for up to two minutes. The medium model, packing 1.4 billion parameters, and the large model, with 2.7 billion parameters, can both create full compositions lasting 6 minutes and 20 seconds.

This represents more than double the capacity of Stable Audio 2.0, released in 2024, which was limited to shorter clips. The ability to sustain musical structure over six minutes opens up possibilities for AI-assisted songwriting and production, potentially aiding musicians in sketching out ideas or creating backing tracks.

Open weights and licensing strategy

In a nod to the open-source community, Stability AI is making the small SFX, small, and medium models available with open weights. This means developers and researchers can download, use, and modify these models freely, fostering innovation and experimentation. The decision aligns with the company's history of open releases, such as Stable Audio Open in 2024, which allowed for 47-second music generation.

However, the large model is positioned as a commercial product. It is accessible only through the company's API and self-hosting paid services, targeting professional and enterprise users. Furthermore, companies with annual revenues exceeding $1 million must secure an enterprise license, adding a layer of control over high-value deployments.

Competitive landscape and legal challenges

The AI music generation field is becoming increasingly crowded and litigious. Stability AI faces competition from well-funded firms like Google and ElevenLabs, each investing in audio AI tools. Meanwhile, startups such as Suno and Udio are embroiled in court battles over the licensing of training data, highlighting the legal uncertainties in the sector.

These disputes underscore the importance of securing proper licenses for training data. As Suno and Udio's cases prove, unauthorized use of copyrighted material could jeopardize the long-term survival of such services. Stability AI appears to be proactively addressing this by partnering with major music labels.

Partnerships and licensed data

To bolster its legal standing, Stability AI has forged partnerships with Warner Music Group and Universal Music Group. These deals involve developing models and music-creation tools in collaboration with the labels, ensuring that the training data is properly licensed. The company emphasizes that its latest audio models are built on fully licensed data, a strategic move to avoid the pitfalls encountered by some rivals.

Such partnerships not only provide legal cover but also integrate AI capabilities with industry expertise, potentially leading to more musically coherent outputs. By aligning with established players, Stability AI aims to position itself as a responsible innovator in the music tech space.

Professional musician tools and executive hires

Beyond the model releases, Stability AI is developing a new suite of products specifically for professional musicians. While details are scarce, the company hints at features tailored to advanced users. To lead this effort, Ethan Kaplan, former chief digital officer at Universal Audio and Fender, has joined the company. Kaplan's experience in music technology suggests a focus on practical, high-quality tools for artists.

This hiring trend reflects a broader industry pattern where AI companies recruit music executives to bridge the gap between technology and creative workflows. Earlier this year, Suno hired Jeremy Sirota, former CEO of Merlin, as chief commercial officer, and ElevenLabs brought on Derek Cournoyer from Kobalt as a strategy lead for music. These moves indicate a growing recognition that success in AI music requires deep industry ties.

Industry trends and future outlook

The release of Stability Audio 3.0 signals a maturing AI music generation landscape, where longer, more complex compositions become feasible. As models improve, they could democratize music production, allowing amateurs and professionals alike to generate high-quality audio quickly. However, challenges around copyright, monetization, and artistic authenticity remain unresolved.

Looking ahead, the integration of AI tools into standard music software and workflows seems likely. Artists may increasingly use AI for inspiration or to handle repetitive tasks, while labels and publishers will seek to monetize these new capabilities. Stability AI's balanced approach of open models and commercial licensing might serve as a blueprint for others navigating this evolving space.