Sonilo taps Musical AI to track licensed music in AI-generated outputs

Sonilo has partnered with Canadian rights-technology company Musical AI to track how licensed recordings and compositions contribute to the music generated by Sonilo’s AI system. The companies said the arrangement is intended to give artists, songwriters and rights holders attribution as AI-generated music becomes more widely used.

Musical AI will analyse Sonilo’s generated outputs and measure the contribution of licensed works to each one. The resulting reports are meant to give Sonilo’s licensing partners usage data that can support compensation for the creators and rights holders behind the source music.

The deal tackles a problem that licensing alone does not resolve: permission to use a catalog does not by itself identify the role individual recordings and compositions played in a particular generated work. Sonilo and Musical AI said they aimed to build licensing, attribution and compensation into the generative-music process from the outset.

Sonilo generates music and sound effects from video footage, text prompts or both, using models built around licensed source material. Its San Francisco-based business is backed by B Capital. Musical AI provides consent-management and attribution infrastructure spanning composition and sound-recording rights.

The announcement came while Sonilo was finalising master-recording and publishing agreements with major rights organisations and preparing an update to its music model. Keli Li, Sonilo’s co-founder and chief operating officer, said the industry needed a better way to connect AI-generated music to the artists and rights holders whose work contributed to it. Musical AI chief executive Sean Power said rights holders needed visibility into that contribution.

Sonilo has previously sought licensed inputs for its models. In May, it said it had licensed Shutterstock’s music catalog for AI-model training, in what Shutterstock described as its first partnership with a video-to-music AI platform. Sonilo’s v1.0 model takes video as an input and composes a soundtrack designed to match its pacing, structure and emotional arc.

The partnership arrives amid broader concern over whether existing music remuneration systems can handle generative AI. An academic paper said piecemeal licensing arrangements lacked scalability and that practical attribution mechanisms often produced uncertain estimates. It distinguished attribution of training data from attribution tied to a specific output, arguing that the latter can enable more direct and verifiable compensation when a catalog is used to condition generation.

Sonilo’s agreement does not specify whether its reporting follows that research framework. But it puts output-level analysis at the centre of a commercial licensing strategy as the company prepares its model update.