A framework for authorship, rights and distribution in the age of AI music. Because a track is never just "AI or human", and its rights follow from every step that made it.
Position measures authorship. Colour measures clearance. The two are independent, and confusing them is the most common error in the AI music debate.
The public conversation about AI music is stuck on a false binary: "real" music versus "AI" music. In practice, creation runs along a continuous spectrum, from fully machine-generated to fully human-performed, and almost everything interesting happens in between. A finished track, moreover, is rarely the product of a single action: it is typically the outcome of many steps (generate, edit, layer, swap, extend) performed through a graphical studio, a conversational AI agent, or both, and its rights follow from that entire chain. This paper introduces the Spectrum of Creation: a shared vocabulary of creation modes, a four-axis model of musical provenance, a three-tier clearance system, and a single composition rule, the weakest-link rule, that together answer three questions the industry has struggled to answer consistently: who made this, can it be released, and where does the value belong?
Between "fully AI" and "fully human" lie dozens of distinct creation modes. Each gets a stable code, a name and a definition.
Authorship and clearance are orthogonal axes. A human song on an uncleared sample is unreleasable; a clean fully AI song is not.
Lyrics, composition, voice, source audio. An uncleared third-party right on any axis, a taint, is what blocks a track.
A track's clearance is the minimum across every step in its lineage. One tainted step taints the chain; no later step can cure it.
Human expression is ownable; pure machine output often is not. The more you write, perform and shape, the more the work is yours.
The situations that cause the most real-world confusion, each paired with the correct analysis and the cure.
No third-party taint on any axis. Built from AI output, your own original input, and in-platform generation. One-click eligible.
Releasable with a condition attached: a licensed artist model with its royalty (tracked by TRACE), a documented consent, or an attestation.
An uncleared third-party right is present. You see exactly which input caused it and how to cure it: remove, replace, or license.
Final clearance = the minimum of every step in the lineage. Original lyrics do not cure a borrowed melody, and editing never launders. Whether the steps happen in the studio or through a chat with Agent One, the ledger is the same.
Lyrics and composition are separate rights. Original words do not cure a borrowed melody. The cure: extract abstract style instead, and generate fresh.
Performing an input does not make it yours. A protected melody stays protected whether it arrives by keyboard or by voice.
A voice can be infringed without a note being copied. Personality and publicity rights protect it independently of copyright.
A fresh composition steered at a living persona raises soundalike and publicity concerns. Keep style descriptors generic.
Ten more hard cases inside the paper, each with the analysis and the cure.
Human expression is ownable; pure machine output often is not. So the creator's share of streaming value rises with contribution, up to a 90% creator share on fully human work. Illustrative creator share of the rights pool shown; exact terms live in the Terms of Service.
Consent-trained artists are paid on every use of their likeness, logged per generation by TRACE. Write the hook, hum the line, sing the verse: every step up the spectrum is a step up in ownership.
Twelve pages: the creation-mode codex, the four provenance axes, all fourteen hard cases and the weakest-link proofs. Scroll through it below, or open it full-screen.
Name the modes precisely. Keep authorship and clearance orthogonal. Compose rights by the weakest link. Pay likeness on every use. Reward the human contribution.