wikiloops is built on one simple promise: every track here was written and played by the person who uploaded it. Our public license only works because of that promise - when you upload, you declare that you are the author of the music and that no third party holds rights in it. It is the same reason cover songs are not allowed.

Music generated by an AI service breaks that promise in several ways at once.

You cannot grant us rights you do not hold.
Under German copyright law - and in most other countries - copyright protects a personal intellectual creation by a human being. Output produced by a generative model is not that. Nobody authored it in the legal sense, so there is no copyright in it for you to license to us, and none for us to pass on to anyone downloading it. A track without a human author cannot be carried by the wikiloops public license at all.

A service's terms are not the same thing as copyright.
Some AI services promise in their terms that you "own" what you generate. That is a contractual promise from a company, not authorship. It comes with conditions, and it changes when the terms change. Our license needs a clean line running from one identifiable human being to wikiloops, and a terms-of-service page cannot supply that.

The authorship becomes shared in a way we cannot map.
wikiloops credits the musicians who played on a track, and anything we ever do with these recordings is built on being able to identify them. A generative service is a participant in the chain that cannot be credited, cannot be identified and cannot consent. The whole structure assumes one author and performer behind each part.

And it carries the same risk we avoid with covers.
Generative models are trained on other people's recordings, and their output can resemble or reproduce protected material. Nobody can tell you reliably whether a generated piece contains something belonging to somebody else. Avoiding exactly that uncertainty is why wikiloops has been originals-only from the start.

None of this is about disliking technology. It is about every file in this collection having a human author behind it - that is the whole basis on which we can hand these recordings to anybody who wants to play along.

Where the line runs: creating music vs. producing it

One distinction decides almost every case: did the tool create musical content, or did it work on musical content that you created?

That distinction is not something we invented for AI. Production has never been part of authorship. An engineer who mixes your song, a mastering studio that brings it up to level, a plugin that removes hiss, a compressor, an EQ, a reverb, pitch correction, timing alignment - none of these make anybody a co-author of the composition or of the performance. They shape a recording of music somebody else wrote and played. Putting machine learning inside a mastering chain does not change that; it is a cleverer tool doing a job that tools have always done. Mastering your take with an AI-backed assistant, cleaning up hiss or hum, nudging a note into tune or a beat into place - all of that is production, and all of it is fine here.

The other direction is the problem. If a tool invents notes, chords, a melody, a bassline, a drum groove, lyrics, or a sung or spoken performance, then musical content came from somewhere other than you - and that is what the declaration on the upload page rules out. A song written from a prompt, a generated backing part, an "AI bandmate" that plays along with you, a synthetic voice singing your lyrics: in each of those, something other than a musician made a musical decision.

Worth saying plainly, because it worries people: a sampled piano, a modelled guitar amp, a drum sampler or a synthesizer is an instrument you play. You decide every note. Those have always been fine here and nothing about that has changed.

Tools on your computer vs. services on the internet

Even when a tool sits clearly on the production side of that line, it matters a great deal whether it runs on your own machine or on somebody else's.

A plugin installed on your computer is the clear case.
Your audio never leaves your studio, nobody else's terms attach to the result, and nobody acquires anything by processing it. Whatever the plugin does internally, what comes out is your recording, processed. There is nothing to sort out.

An online service is a different situation, and it is worth taking seriously. When you upload audio to a web service, you hand a copy of a recording to a company, and their terms decide what happens next: what they may do with what you sent, whether they keep it, whether they use it to train their systems, and what rights - if any - attach to what comes back to you. Some services claim a broad license in everything uploaded to them. Others are perfectly reasonable. You cannot tell without reading, and the answer differs from one service to the next, and from one version of the terms to the next.

For your own recordings, that is your decision to make, and we are not going to tell you which services to use. Please just be aware that it is a decision - "the result sounded great" is a different question from "what did I agree to when I uploaded my stem".

For recordings that are not yours, it is not your decision to make at all.

Please do not feed wikiloops tracks into AI services

Everything you download from wikiloops comes to you under the downloads license, a grant for private use. That grant is ours to give because our members gave us the right to pass it on - and it is deliberately narrow. The article on the wikiloops public license explains how that chain works.

Uploading a wikiloops track to a third-party AI service is not reliably covered by it. The moment the file leaves your computer for someone else's servers, that company's terms apply to it, and some of those terms include a license to use whatever is uploaded - to improve a service, to train a model, or simply to retain it. You would be granting rights in another musician's recording that were never yours to grant, and that the musician never agreed to.

This applies whether the service is a song generator, a mastering service, a stem separator or anything else, and it applies whether or not you intend to upload the result back to wikiloops.

If you want to work on a wikiloops track - remix it, separate it, practise along with it - please use tools that run locally. The line here is not the technique. It is where the audio goes.

Tell people in your track description

If you experimented with an AI-backed tool somewhere in your production - a mastering assistant, a separation tool, a denoiser - the community would genuinely appreciate a line about it in your track description. It is not a rule breach and nobody will think less of you for it. People here are curious about what these tools actually do, and a note on what you used and how it went is interesting reading.

And if you are unsure whether something you would like to try falls on the right side of the line, please ask in the forum before you upload. We would much rather answer a question than have you find out afterwards.

This wikiloops