Stability AI Releases Stability Audio 3.0 for Generating Multi-Minute Songs

Stability AI introduced an updated version of its audio model, Stability Audio 3.0, which can generate musical compositions up to six minutes long. This is a significant step forward in the development of AI-based music synthesis systems.

Two Versions for Different Tasks

The company released the model in two variants, each targeting different use cases. The full-featured version creates music up to 6 minutes long — these are complete songs, not just snippets. The lightweight version, called the small model, runs directly on user devices without requiring connection to cloud services.

Local operation of the model offers several practical advantages. First, there is less latency in generation — results are produced almost instantly. Second, there is greater privacy — no data is sent to company servers. Third, users are not dependent on internet connection stability.

Capabilities and Applications

Stability Audio 3.0 continues to develop the text-to-audio approach that Stability AI initiated with previous versions of its model. The system understands natural language descriptions and converts them into music of the desired style and duration.

Key capabilities include:

• Generating original compositions based on text descriptions of genre, mood, and style
• Support for various music genres from classical to electronic
• Working with long and detailed descriptions for precise generation of the desired sound
• Local execution on personal computers without cloud infrastructure
• Scalability from short 20-second clips to full six-minute compositions

These capabilities open the way for practical applications across various fields.

Where This Can Be Useful

For musicians and content creators, Stability Audio 3.0 can become a tool for rapid prototyping of musical ideas. Instead of spending time searching for ready-made tracks or collaborating with composers, they can generate multiple variations of background music, video soundtracks, or demo versions of their compositions in minutes.

For independent streamers and video bloggers, local music generation solves the copyright issue — they can create original compositions without purchasing licenses for pre-made music.

What This Means

The market for generative audio models is becoming increasingly competitive and accessible. While music synthesis was previously controlled by a few large companies with cloud infrastructure, open models are now emerging that work locally and do not require significant computational resources.

This democratizes music creation and removes dependence on cloud services and subscriptions.