Stability AI has launched Stable Audio 3.0, featuring three open-weight models capable of generating music tracks up to six minutes long. All models were trained exclusively on licensed data.
Stability AI's latest audio generation platform represents a significant advancement in music synthesis capabilities. The new Stable Audio 3.0 suite extends generation length to six minutes per track, up from previous limitations, enabling creators to produce longer-form compositions in a single generation.
Three of the released models feature open weights, meaning developers can access and modify the underlying architecture. This approach aligns with Stability AI's broader strategy of democratizing AI tools while maintaining commercial viability through different licensing tiers.
The company emphasizes that all training data used for Stable Audio 3.0 came from licensed sources. This addresses ongoing industry concerns about copyright and fair compensation for artists whose work may be used in AI training datasets. By utilizing licensed material, Stability AI positions the models as legally defensible solutions for commercial use.
Stable Audio represents the company's expansion beyond its flagship image generation tool, Stable Diffusion. The audio models join a growing competitive landscape that includes platforms like OpenAI's Jukebox, Google's MusicLM, and other emerging generative audio tools.
The six-minute capability targets practical use cases including background music composition, podcast intros, and short-form content creation. Users can generate tracks across multiple genres and styles without requiring external audio composition tools.
Stability AI has not disclosed specific performance benchmarks or comparative data against competing platforms. The open-weight approach may accelerate community innovation, though model size and computational requirements for local deployment remain undisclosed.
The release follows Stability AI's recent financial challenges and restructuring, signaling continued commitment to model development despite the broader AI industry's consolidation pressures. Availability and pricing details for different access tiers have not been fully detailed.
Google has launched 'Expert Intelligence,' a new feature in Gemini Notebook that pulls content from Google Play Books, allowing users to ask questions and generate AI-assisted materials based on book contents.
Synthetic creators are building political communities ahead of Brazil's presidential election, using AI-generated personas to engage voters around specific political views.
Australia's recording association has prohibited fully AI-generated music from official charts, though tracks using AI as a production tool remain eligible if substantially created by humans.
Neurosurgeons at a London hospital have successfully completed the world's first AI-assisted operation to remove a brain tumor. The procedure, performed in May, preserved the vision of a 48-year-old patient.