Stability AI Unveils Stable Audio 2.5, the First Audio Model for Enterprise-Scale Sound Production

Stability AI has launched Stable Audio 2.5, the first audio generation model designed for enterprise-scale sound production. This model helps brands consistently apply their unique sound identity across various channels.

Key improvements include the following:

• Fast generation speed: Through the ARC (Adversarial Relativistic-Contrastive) method, a 3-minute-long track can be generated in under 2 seconds on a GPU.
• Intelligent composition: It understands the structure of music (intro, development, ending), and its ability to follow prompts such as mood or instrument descriptions has been improved.
• Audio Inpainting: It supports a feature that grasps the context of user-input audio and naturally generates the remaining parts.

Stable Audio 2.5 was trained on a copyright-secured, licensed dataset, making it safe for commercial use. It also supports Fine-tuning using a company's own sound library, helping build a brand's unique sonic identity.

The model is currently available through the Stability AI API, fal, Replicate, and ComfyUI, with on-premises deployment and custom solutions also offered through enterprise licensing.