The science of musical influence.

Generative AI learns from music. LUMINA™ measures which music influences what the model creates, how, and by how much.

Developed by Fold Artists, LUMINA™ uses proprietary, patent-pending technology to measure the influence of individual training works on specific AI-generated outputs. It gives AI platforms (licensees) and music rightsholders (licensors) a new way to understand the contribution of music within generative AI systems.

Influence is more than resemblance.

Audio similarity

Can tell us what two recordings have in common.

Training-data influence

Requires examining how the AI model has learned from the human material it was trained on.

Two recordings can sound similar without having the same influence on an AI model. And a work that shaped the model's learning may leave little obvious musical resemblance in the finished output. That distinction matters.

LUMINA™ approaches attribution from inside the generative model, measuring influence through the model's internal learning signals rather than relying on comparisons between finished recordings alone.

From human training data to AI-generated music.

Every generative AI model is shaped by its training data. But the presence of a work in that data does not, by itself, tell us how much it influenced a particular output.

By using proprietary technical methods to measure the relationship between training works and generated results at the model level, LUMINA™ provides influence measurements that can be analyzed, compared and incorporated into technical and commercial reporting.

Conceptual illustration: a field of training works, a few of which are connected by lines of light to a single generated output, with line strength suggesting how much each work influenced it. Not a depiction of the method.

Which musical works influenced this output, and by how much?

Built for the complexity of music.

Music is more than a single creative asset. A recording brings together composition, performance, production and sound, and a generative model learns from all of them at once.

So LUMINA™ does not split influence into separate composition and recording scores. It produces a single ranking of the training works that shaped each output, grounded in the operation of the AI model itself.

One output. One ranking of what shaped it.

A foundation for a licensed AI music economy.

AI platform operators (licensees)

Measure the influence of training works on generated outputs. Build a stronger and fairer technical foundation for licensing relationships, rightsholder reporting and responsible commercial deployment.

Music rights owners (licensors)

Gain new visibility into how repertoire contributes to generative AI music. Use influence measurements to inform licensing negotiations, as well as fair distribution of incoming royalties amongst your own upstream stakeholders.

The wider music industry

Connect influence calculations with the infrastructure that manages rights, metadata, distribution and payments.

Licensing establishes the commercial terms. LUMINA™ provides measurements that can inform how they are implemented.

Built on mathematical research.

LUMINA™ combines proprietary research with an ongoing program of technical development and evaluation.

Our work includes controlled experiments and confidential industry testing, with a focus on reproducibility, measurement quality and the requirements of larger-scale B2B deployments.

Understand what shaped the AI output.

We're engaging with AI developers, music rightsholders, digital music distributors, and adjacent technology partners interested in evaluating LUMINA™.