UK-based Stability AI has unveiled a new audio generation model called Stable Audio 2.5, which the artificial intelligence company says is tailored for enterprise-grade sound production.
The new model addresses challenges in corporate marketing. Stability noted, citing Ipsos research, that while custom audio can make brands eight times more memorable, only 6% of creative campaigns incorporate a sound identity.
The company’s research suggests that 86% of brand engagement is influenced by audio, but only few brands use custom audio.
Stability AI says Stable Audio 2.5 provides “faster generation, smarter composition, enhanced workflows,” capable of generating three-minute compositions in under two seconds on H100 GPUs.
The model was post-trained using what Stability AI calls Adversarial Relativistic-Contrastive training, a method developed by the company’s Stable Audio research team.
According to Stability AI, the new version has improved musical structure, producing compositions with “richer multi-part compositions,” and is able to respond more effectively to prompts like “uplifting” or specify musical elements such as “lush synthesizers,” suggesting that the system responds to both emotional and technical directions.
A new feature called audio “inpainting” allows users to upload existing audio files and specify where they want additional content generated. The model analyzes the context and extends the track accordingly.
Stability AI, maker of the popular Stable Diffusion image generator, emphasized that its terms of service prohibit copyrighted material uploads, with content recognition systems monitoring compliance.
Its statement comes as Stability AI faces multiple copyright infringement lawsuits. In one case, Seattle-based Getty Images sued Stability AI in both the UK and the US, alleging that the company illegally used 12 million images without permission or compensation.
“Like all Stable Audio models, Stable Audio 2.5 is commercially safe and trained on a fully licensed dataset.”
Stability AI
Separately, illustrators Sarah Andersen, Kelly McKernan, and Karla Ortiz filed a class action lawsuit in January 2023 against Stability AI and two other companies, challenging the use of their works for AI model training.
Most recently, Stability AI said: “Like all Stable Audio models, Stable Audio 2.5 is commercially safe and trained on a fully licensed dataset.”
Coinciding with the launch of Stable Audio 2.5, Stability AI announced that it is partnering with Amp, a WPP-owned sound branding agency, to develop enterprise solutions for brands. The collaboration will make Stable Audio 2.5 available to WPP’s global client portfolio through the advertising giant’s technology platform, WPP Open.
Back in July, Stability AI CEO Prem Akkaraju told the Financial Times in an interview that the company is looking at developing a marketplace where artists can license their work for AI training.
The proposed marketplace would allow creators to voluntarily submit artwork and receive compensation when AI companies use their content for model training.
Music Business Worldwide