Stability AI has introduced Stable Audio 2.5, its new enterprise-focused audio generation model designed for corporate sound production. The company highlighted research showing that while custom audio can make brands significantly more memorable, only a small percentage of marketing campaigns use a distinct sound identity. Stable Audio 2.5 is positioned as a solution, offering faster generation, improved composition, and enhanced workflows. It can generate three-minute tracks in under two seconds using H100 GPUs, and was refined with Stability’s proprietary Adversarial Relativistic-Contrastive training method.
The new version emphasizes stronger musical structure, producing layered, multi-part compositions and responding more effectively to descriptive prompts, whether emotional or technical. Features include audio inpainting, which lets users extend existing tracks by generating additional content in chosen sections. Stability AI stressed that its terms prohibit copyrighted uploads, with safeguards in place to monitor compliance.
The company also confirmed that Stable Audio 2.5, like its earlier models, was trained entirely on licensed datasets, underscoring its claim of commercial safety. Alongside the release, Stability AI announced a partnership with Amp, a WPP-owned sound branding agency, to integrate Stable Audio 2.5 into WPP Open, the advertising network’s global technology platform.
Looking ahead, Stability AI is considering a marketplace that would let artists license their work for AI training and be compensated for its use, an idea CEO Prem Akkaraju previously discussed in an interview. This proposed marketplace would mark a step toward voluntary, transparent collaboration between creators and AI developers.
