In March 2026, the U.S. Supreme Court declined to review a ruling that copyrightable works require a human author. The denial did not create a new music rule, but it left the existing human-authorship position intact. Eleven Music can grant broad usage permission without changing that threshold.
ElevenLabs says that, as between you and the company, you retain rights in your output. You may also receive commercial rights attached to Eleven Music tracks, yet a contract cannot create federal copyright in expression produced entirely by a model.
ElevenLabs treats each output as both a sound recording and the musical composition embodied in it. The composition consists of the music and any lyrics, while the recording is the particular fixed performance and production. Human authorship can exist in one work without extending to the other, so each needs its own analysis.
Detailed prompting does not solve the problem. Choosing a genre, tempo, key, instrumentation, mood, and song structure may strongly influence a result, but current guidance says prompts alone do not provide sufficient control over the expressive output. The useful test is whether human-authored musical expression can still be identified in the final track.
A melody you composed and supplied as an expressive input may remain protected when it is clearly perceptible in the generated version. The same principle can apply to your lyrics or other original material, although protection reaches that surviving contribution rather than every sound around it.
Keep the evidence clean. Dated lyric drafts, melody recordings, notation, MIDI exports, session files, and saved generations can show what existed before the model intervened and what you changed afterward. Those records do not manufacture authorship, but they can make a genuine contribution much easier to identify.
Creative arrangement can also matter. Selecting several generated passages, rejecting others, ordering them into a deliberate form, and writing original transitions may support protection for the resulting selection and arrangement. The underlying AI passages remain outside that scope, and a routine trim or volume adjustment may be too slight to qualify.
Production choices may support additional authorship. Choosing takes, chopping stems, rebuilding timing, designing transitions, recording live instruments, and shaping the final sequence can amount to original production or engineering when the choices are sufficiently creative. Protection follows those choices, not the mere act of exporting a file.
There is no official percentage that makes a track human enough. Ten edits are not automatically stronger than three, and spending twenty hours does not prove creative control. The question is whether your decisions formed identifiable expression in the particular recording.
Synthetic vocals need careful treatment. Writing the words may support a composition claim in the lyrics, but it does not make you the performer of a vocal generated entirely by the model. Replacing that vocal with your own recorded performance changes the sound-recording analysis while leaving the AI-generated musical material subject to the same limit.
A careful description might claim original lyrics, an original melody that remains audible, vocal performance, production, or a creative selection and arrangement. The excluded material might include generated music, generated vocals, or untouched generated passages. Precision matters because the registration covers the stated human authorship, not every element stored in the audio file.
Platform ownership language still needs a narrow reading. ElevenLabs warns that outputs may not be unique and may resemble or match material returned to another user, so contractual rights do not promise exclusive expression. A registration limited to your lyrics, vocal performance, and production cannot be stretched later to cover an independently generated melody that your claim expressly excluded.
ElevenLabs says that, as between you and the company, you retain rights in your output. You may also receive commercial rights attached to Eleven Music tracks, yet a contract cannot create federal copyright in expression produced entirely by a model.
ElevenLabs treats each output as both a sound recording and the musical composition embodied in it. The composition consists of the music and any lyrics, while the recording is the particular fixed performance and production. Human authorship can exist in one work without extending to the other, so each needs its own analysis.
Human lyrics can survive an AI-made composition
If you write original lyrics and Eleven Music sets them to generated music, the words can remain your protected contribution. The AI-made melody, harmony, rhythm, and vocal delivery do not automatically join that claim merely because they carry your lyrics.Detailed prompting does not solve the problem. Choosing a genre, tempo, key, instrumentation, mood, and song structure may strongly influence a result, but current guidance says prompts alone do not provide sufficient control over the expressive output. The useful test is whether human-authored musical expression can still be identified in the final track.
A melody you composed and supplied as an expressive input may remain protected when it is clearly perceptible in the generated version. The same principle can apply to your lyrics or other original material, although protection reaches that surviving contribution rather than every sound around it.
Keep the evidence clean. Dated lyric drafts, melody recordings, notation, MIDI exports, session files, and saved generations can show what existed before the model intervened and what you changed afterward. Those records do not manufacture authorship, but they can make a genuine contribution much easier to identify.
Creative arrangement can also matter. Selecting several generated passages, rejecting others, ordering them into a deliberate form, and writing original transitions may support protection for the resulting selection and arrangement. The underlying AI passages remain outside that scope, and a routine trim or volume adjustment may be too slight to qualify.
Human performance creates a separate recording claim
The sound recording analysis starts somewhere else. Copyright in a recording can arise from a performer's contribution and from a producer or engineer who captures, manipulates, or edits the sounds in the final master. A human vocal recorded over an AI-generated backing can therefore create protectable performance authorship without converting the generated backing into human-made music.Production choices may support additional authorship. Choosing takes, chopping stems, rebuilding timing, designing transitions, recording live instruments, and shaping the final sequence can amount to original production or engineering when the choices are sufficiently creative. Protection follows those choices, not the mere act of exporting a file.
There is no official percentage that makes a track human enough. Ten edits are not automatically stronger than three, and spending twenty hours does not prove creative control. The question is whether your decisions formed identifiable expression in the particular recording.
Synthetic vocals need careful treatment. Writing the words may support a composition claim in the lyrics, but it does not make you the performer of a vocal generated entirely by the model. Replacing that vocal with your own recorded performance changes the sound-recording analysis while leaving the AI-generated musical material subject to the same limit.
Accurate registration protects the claim you actually have
A registration application for a work containing more than minimal AI-generated material should identify the human contribution, exclude the generated material, and describe composition and sound-recording authorship separately. One Standard Application can cover both only when the same claimant owns them, and they appear on the same phonorecord. Naming yourself as author of the entire track can overstate what copyright protects.A careful description might claim original lyrics, an original melody that remains audible, vocal performance, production, or a creative selection and arrangement. The excluded material might include generated music, generated vocals, or untouched generated passages. Precision matters because the registration covers the stated human authorship, not every element stored in the audio file.
Platform ownership language still needs a narrow reading. ElevenLabs warns that outputs may not be unique and may resemble or match material returned to another user, so contractual rights do not promise exclusive expression. A registration limited to your lyrics, vocal performance, and production cannot be stretched later to cover an independently generated melody that your claim expressly excluded.
