AI music evidence file for an agency hand-off: job, engine, terms

Keep one evidence record per AI music track in an ad: the vendor and plan, the engine or model id, the date, the prompt, the job or download record, and the version of the terms you read. Sume gives you three of those on the job itself: the engine in job.request.routed_model, caller metadata stored on the job, and the audio artifact URL on media.sume.com. The rest comes from the vendor pages.

What each source gives you

Each system holds part of the proof. Collect the part it can provide and note the date.

Setting it up on Sume

The Music 1.0 page says metadata is caller metadata stored on the job and not sent to the provider. Put there the client code, campaign, brief version and the terms-read date. Pin an engine such as lyria-3.5 instead of sume/music-auto when the log must name a fixed engine; otherwise read job.request.routed_model after completion, as described in the Music Router docs. Save the artifact URL and, if you need it later, copy the file to your own storage.

Why vendors need separate rows

Terms are not the same across vendors. Suno's terms gate commercial use on plan and download, ElevenLabs' music terms restrict some sectors and ban some prompt inputs, and say output may not be unique, and neither is a statement about Sume. Do not paste one vendor's clearance into another vendor's row.

• Record the date you read each page; they change.
• Keep a copy of the clause, not only a link.
• Store the exact prompt used for the delivered take.
• Note who approved the track and when.

The hand-off

When the job closes, hand the client a one-page record and the audio file. If a platform queries the music later, you can answer from the record instead of reconstructing it. See Jobs and results for how to fetch a finished job again. This is a process suggestion, not legal advice.