Stability AI launches Stable Audio 3.0 with up to six-minute tracks and open weights

2026-05-20 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Stability AI has launched Stable Audio 3.0, a new generation of audio models, with three of its four variants available as open-weights. These models generate music tracks up to six minutes long and were trained entirely on licensed data. The family includes Stable Audio 3.0 Small SFX and Small (459 million parameters, 2-minute tracks, 0.44 seconds inference on an H200 GPU), and Medium (1.4 billion parameters, 6:20 minute tracks, 1.31 seconds inference). The largest model, Stable Audio 3.0 Large (2.7 billion parameters), is exclusive to API users and enterprise customers. The new architecture features a semantic-acoustic autoencoder for flexible output and inpainting capabilities. Stability AI emphasizes its licensed training data and offers legal indemnification for enterprise users, positioning itself against competitors facing copyright lawsuits, such as those involving OpenAI, Suno, and Udio. Commercial use is free for organizations with under \$1 million in annual revenue.

Key takeaway

For AI Engineers or Music Producers evaluating generative audio tools, Stable Audio 3.0 offers a compelling, legally safer option. You can utilize its open-weight models for commercial projects up to \$1 million in revenue without licensing fees. Consider its licensed training data and enterprise indemnification as a critical differentiator, mitigating copyright infringement risks prevalent with other platforms. Explore the LoRA fine-tuning capabilities to customize models with your own audio libraries.

Key insights

Stability AI's new audio models prioritize licensed data and legal indemnification to mitigate copyright risks.

Principles

• Licensed data reduces legal exposure.
• Open weights foster broader adoption.
• Enterprise indemnification adds value.

Method

The new semantic-acoustic autoencoder architecture enables variable-length audio generation with second-level control and inpainting features for editing and extending tracks.

In practice

• Fine-tune models using LoRA documentation.
• Generate sound effects on mobile devices.
• Edit specific segments of generated audio.

Topics

• Stable Audio 3.0
• Generative Audio
• Open Weights
• Copyright Compliance
• AI Music Generation
• Enterprise AI Licensing

Best for: CTO, Machine Learning Engineer, AI Product Manager, AI Engineer, Director of AI/ML, Legal Professional

Related on AIssential

• Warner Music Group and Stability AI Join Forces To Build The Next Generation Of Responsible AI Tools For Music Creation — WMG and Stability AI partner to develop ethical, artist-centric generative AI tools for music creation.
• Lyria 3 Pro: Create longer tracks in more Google products — Lyria 3 Pro extends AI music generation to 3-minute tracks with structural control across Google's ecosystem.
• The Open-Source Bet on AI Video Nobody's Talking About — Open-source AI video models with integrated audio-video generation offer a compelling alternative to closed, high-cost platforms.
• Recursive Learning Without Collapse: A Weighting-Based Stabilization Framework — Optimal weighting in recursive generative model training prevents model collapse by balancing real and synthetic data.
• The Atlantic created a searchable database of the music used to train AI — AI music models are trained on vast, publicly available datasets, often acquired through methods that violate platform terms of service.
• Real-Time Interactive Music Generation via Data-Free Streaming Consistency Distillation — A novel framework transforms static generative music AI into real-time, interactive instruments for live co-creation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.