Two plugins in one chain can disclose their training differently

The audio task

A producer chaining two AI-assisted tools together, a generative sketch tool and a separation or mastering plugin, may find that each vendor describes its own training data differently. One might name specific licensed sources; another might describe its dataset only in general terms, or focus its published terms on usage tiers rather than data provenance at all. Deciding whether that gap matters for a given release is a judgment call, not something either vendor's page resolves.

What the documents show

Stability AI's licence page describes the Stability AI Community License by usage and revenue tier, stating it 'allows for research, non-commercial, and commercial use of the Core Models for individuals or organizations that generate under $1M... of annual revenue,' with a paid Enterprise License required above that threshold; the page does not name specific training data sources. Meta's MusicGen model card, by contrast, states plainly that the model 'was trained on licensed data using the following sources: the Meta Music Initiative Sound Collection, Shutterstock music collection and the Pond5 music collection.'

Rights status

Neither page's disclosure level is asserted here to be false or inadequate; Stability's licence page addresses who may use its models commercially and at what scale, while MusicGen's model card addresses what data trained the model. They answer different questions, published in different document types, and a producer should not read a revenue-tier licence as a data disclosure or the reverse. Where a specific project needs training-data provenance and a vendor's licence page does not supply it, that is a gap to note, not evidence the vendor is misrepresenting anything.

What to check before you use it

This is an editorial method record comparing named vendors' own disclosure language, not a resolution of which vendor is correct. When two tools in a chain differ this way, check each vendor's full documentation, not just the page most prominently linked, since a training-data statement may live in a separate model card, FAQ, or terms page from the licence itself; and record which disclosure, if any, backs each tool in the session log.

• Does either vendor's documentation name specific training sources, or only describe usage terms and revenue tiers?
• Is the gap between the two disclosures relevant to this specific release, or only a general difference in document type?
• Would a client or distributor asking about training data be satisfied by what each vendor has actually published?

A licence page and a model card are not interchangeable documents; comparing them fairly means asking what each one was written to disclose, before treating a gap between them as a problem.