Four sources, with what each actually permits
Commercially licensed libraries. Artlist, Epidemic Sound, Musicbed and similar. You pay a subscription and get a licence covering specified uses. Read what it covers: many subscriptions license content published while you are subscribed and continue to cover it afterwards, some do not, and most treat broadcast and paid advertising differently from social posting. This is the safe route for client work and it costs money.
Creative Commons. Free Music Archive, ccMixter, parts of Freesound, and a great deal of independent work. Free, and the licence terms vary per track and matter:
- CC0: no conditions at all.
- CC BY: attribution required, which means actually crediting the artist in the description or on screen.
- CC BY-NC: non-commercial only. A client video is commercial. A monetised channel is commercial. This catches people constantly.
- CC BY-SA: share-alike, which can carry obligations onto your finished work.
Always check the specific track’s licence on the page you downloaded it from, not the site’s general description.
AI-generated music. Suno, Udio and their competitors, plus open models. Fast, specific to your piece, and legally unsettled: see below.
Music you make. Free open tools cover this: LMMS for composition, Audacity for arrangement of loops, Ardour for a full multitrack session. If you play anything, a phone recording of it has no licence questions at all.
The unsettled ground, described rather than resolved
AI music generation sits in the middle of a live dispute and it would be dishonest to tell you otherwise.
The disagreement has two parts. Training: whether using copyrighted recordings to train a generative model requires a licence. Major labels have brought actions against AI music companies; some have been settled or restructured into licensing arrangements, others continue. The answer differs by jurisdiction and no jurisdiction has produced a settled body of law.
Output ownership: whether what the model produces is protectable, and who by. In some countries a work requires human authorship to attract copyright, which puts a purely prompted output on uncertain footing. Platform terms of service often purport to assign output rights to the user, which is a contract between you and that company, and is not the same thing as copyright existing.
What you can do about it practically:
- Read the specific service’s terms for what it says about commercial use, and keep a copy of what it said on the day you generated.
- Prefer models and services that state their training sources, where any do.
- On client work, say in writing which music is AI-generated, so the client is making the decision with you rather than discovering it later.
- Where the stakes are high (broadcast, a paid campaign, anything with real exposure) licensed library music is the boring choice and it is the right one.
This is a description of a disagreement, not advice about your situation, and the position in your country may differ from all of the above.
Platform detection is a separate problem
Independently of whether you have a valid licence, automated content-identification systems on the large platforms may flag your upload. A legitimately licensed track can be claimed because a rights database contains it; an AI-generated track can be flagged for resembling something.
Keep the licence documentation (the receipt, the licence page, the track page) filed with the project. A dispute is resolved by producing it, and the process is much easier when the file is in the project folder rather than in an inbox from eight months ago.
Using music well, which is mostly using less of it
- Start it late. Music from frame one is a common amateur signature. Let the piece begin with ambience and bring music in at the first real beat of the story.
- Duck it under speech. Lower the music where anyone talks. A keyframed level change is enough; automatic ducking exists in Resolve’s free tier.
- Cut picture to the music or music to the picture, and decide which. Doing neither produces the characteristic feeling that the two are unrelated.
- Silence is a tool once the bed exists. Pulling the music out at a moment is a strong effect, and it is available to you precisely because the ambience is still there underneath.
Today: find the licence text for whatever music you last used and read the commercial-use clause. Then file it in the project folder.
Every lesson here is open. Your progress is recorded once you pass the Editing Video exam.
Check your understanding
A freelancer uses a CC BY-NC track in a video for a paying client, crediting the artist in the description. What is the problem?
Pick the one you would defend. Your answer stays private.
Stuck on something?Ask Chiamaka, your AI tutor
