Stability AI Brings AI Audio to Brand Production

Stability AI wants brands to stop hunting for stock music and start generating their own.

Its new Stable Audio 2.5 model can create tracks up to three minutes long, support audio inpainting, and run through WPP’s global client network.

For agencies, marketers, and creative teams, the goal is faster custom audio for ads, games, short-form video, retail experiences, and other campaign work.

From prompts to campaign audio

According to Stability AI’s announcement, Stable Audio 2.5 is built for enterprise sound production and can generate tracks up to 3 minutes long in under 2 seconds of GPU inference time.

Stability AI also said the model was post-trained using a method it calls Adversarial Relativistic-Contrastive training, which is designed to improve quality and speed. The company said Stable Audio 2.5 responds to mood and genre prompts, including terms such as “uplifting” or “lush synthesizers.”

The older Stable Audio Open model was more limited. In 2024, TechCrunch reported that it could generate recordings up to 47 seconds long and was not meant to create full songs, melodies, or vocals. Stability AI’s own Stable Audio Open release also described that model as a tool for short audio samples, sound effects, and production elements.

Stable Audio 2.5 is designed for creative teams who need music and sound across multiple formats. Stability AI said the model is available through its API, partner platforms including fal, Replicate, and ComfyUI, and on-premises through an enterprise license.

The WPP connection explains the intended buyer. Stability AI said it is partnering with amp, a sound branding agency within WPP’s Landor Group, to co-develop enterprise audio solutions. Stable Audio 2.5 will also be available to WPP’s global client base through WPP Open.

For a brand team, that could mean generating different versions of a campaign soundtrack, refining a short audio cue, or adapting sound for several channels without starting from scratch each time.

What brands need to check

Audio inpainting is one of the more practical additions. Stability AI’s prompt guide says users can upload an audio clip, choose a start point or time range, and have the model complete the composition using surrounding context. The company’s terms require uploads to be free of copyrighted material, and Stability AI says it uses content recognition to help prevent infringement.

Stability AI describes Stable Audio 2.5 as commercially safe and trained on a fully licensed dataset. For brand teams, that claim affects whether a generated track can move through legal review, campaign approval, and public distribution without creating new copyright concerns.

The copyright backdrop is still active. The UK has been weighing stricter rules around AI training and creative rights, with lawmakers pressing for more transparency and licensing protections for creators.

Stable Audio 2.5 also arrives as audio and voice become a larger part of the AI product race, from AI assistants to rumored audio-first AI devices.

For creative teams, the practical change is speed and variation. A brand that needs music for a product video, a seasonal ad, and several social cuts could quickly generate options, then refine the results through inpainting or a custom model built around its own sound library.

That does not remove the need for composers, audio directors, licensing checks, or human review. It gives agencies another way to produce and adapt sound across campaign formats while keeping licensing, provenance, and approval questions in the production process.

Also read: OpenAI’s latest voice AI update shows how quickly companies are turning voice tools into systems that can listen, respond, translate, and act.

The post Stability AI Brings AI Audio to Brand Production appeared first on eWEEK.