Stability AI launches Stable Audio 3.0 with up to six-minute tracks and open weights
Key Takeaways
- Three of four model variants in Stable Audio 3.0 are available as open-weights models, allowing users to train their own versions.
- The largest model remains exclusive for API use and enterprise customers due to the need for legal indemnification against copyright claims.
- Stability AI has chosen not to provide fully licensed training data with open weights in order to avoid potential legal disputes, positioning itself as a neutral provider in this space.
Stability AI has introduced Stable Audio 3.0, a new generation of audio models that can generate music tracks up to six minutes long using only licensed training data. This is a significant advancement for the company, which previously relied on partnerships with external providers like AudioSparx.
The model family includes four variants: Stable Audio 3.0 Small SFX and Medium, both designed for shorter compositions, and the larger Stable Audio 3.0 Large. The two smaller models are available as open-weights models, meaning users can fine-tune them further to better suit their needs.
Stability AI’s decision to provide these variants with open weights is a strategic move aimed at ensuring that its tools remain accessible and adaptable for various applications without the risk of copyright violations. This approach aligns with Stability AI’s commitment to being a reliable provider in an industry where such issues can be contentious.
New Architecture Enables Longer, More Flexible Audio Output
The new architecture used by Stable Audio 3.0 allows for longer and more flexible audio outputs compared to previous iterations. This is made possible through a semantic-acoustic autoencoder, which enables the model to generate music tracks of varying lengths without sacrificing quality.
Stable Audio 3.0 Small is particularly noteworthy as it can produce full-length compositions on-device, offering users greater flexibility and control over their audio outputs. This feature sets it apart from other models in its class, providing a more seamless user experience for those looking to use these tools without the need for external processing.
Enterprise Customers Get Legal Protection
To support enterprise customers who might be using Stable Audio 3.0 extensively within their operations, Stability AI has introduced comprehensive legal protections. These include commercial coverage and indemnification from copyright claims related to the use of fully licensed training data.
This move is in stark contrast to competitors like Suno and Udio, which have faced significant legal challenges over the unlicensed nature of their training data. By choosing a path that avoids such controversies, Stability AI positions itself as a more stable and trustworthy provider for its enterprise clients.
From Image to Audio: The Shift Continues
Stability AI has evolved from a pioneer in the image generation space with Stable Diffusion to becoming an audio specialist. This shift is evident not only in their new models but also in how they approach licensing and legal issues.
The company’s history underscores its commitment to providing tools that are both innovative and legally sound, positioning it as a reliable choice for those looking to integrate AI-generated music into their operations without the risk of costly legal battles.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
