What is AI-generated music?
AI-generated music is music that’s composed and produced by an artificial intelligence model, rather than recorded by a human musician or sampled from an existing track. The AI generates the audio itself — melody, instrumentation, and (if the track has lyrics) vocals — from a text prompt describing the style, mood, or genre. Nothing is copied, rearranged, or pieced together from existing recordings. The audio is synthesized, start to finish, by the model.
It’s a newer enough category that it regularly gets confused with a few things it isn’t:
- Not a remix. A remix takes an existing recording and alters it — pitch-shifts it, adds effects, layers in new elements. AI-generated music starts from nothing. There’s no original recording underneath it to remix.
- Not a sample library. Sample-based production stitches together pre-recorded loops and one-shots made by human musicians. AI-generated music doesn’t draw on pre-recorded material at all — every note in the output is generated, not selected from a library.
- Not “royalty-free” music in the traditional sense. Traditional royalty-free tracks are still human-composed and human-performed; they’re just licensed under more permissive terms than a typical commercial recording. AI-generated tracks have no human composer or performer behind the specific recording at all.
How does it actually work?
An AI music model is trained to associate descriptions of sound — genre, instrumentation, era, mood, tempo — with audio patterns. When you give it a prompt like “ambient, warm analog synths, slow tempo,” it generates a new audio file that matches that description. The model isn’t retrieving or recombining stored audio clips; it’s producing a new waveform based on patterns it learned during training, the same general idea as a text model generating a new sentence rather than quoting one it memorized.
This matters for the practical side of licensing and rights, too: because the output isn’t derived from a specific existing recording, it doesn’t carry that recording’s copyright with it. That’s a structurally different situation from, say, licensing a song to play in a shop, where the recording, the composition, and the performance are all somebody else’s copyrighted work — see Can businesses legally use AI-generated background music? for what that distinction actually means for a shop’s licensing obligations.
How FeelFlow Radio uses this
FeelFlow Radio generates a station per listener rather than streaming from a fixed catalogue. When you set up your station, you choose your genres (from 21 available), your language (from 17, including instrumental-only), and either a mood or a live heart-rate reading. Every track that plays is generated to match that specific combination — it isn’t pulled from a pre-built playlist that happens to fit the tags. (For how those choices actually turn into a station, see how does AI radio work?.)
That has a few consequences that are worth spelling out plainly, because they’re easy to gloss over:
No listener ever hears the same track twice. This isn’t a soft goal — it’s enforced at the database level with a uniqueness constraint, so it can’t quietly fail even if there’s a bug elsewhere in the system. When two listeners request the same combination of genre, mood, and language, a newly generated track can be shared between them the first time it’s created — so the generation cost is paid once, not once per listener — but no individual listener is ever served a repeat.
The catalogue effectively has no ceiling. A curated background-music playlist, however good, has a fixed length — and after a few hours on loop in a shop or café, staff working an eight-hour shift will hear it. Because FeelFlow Radio generates new tracks on demand rather than looping a fixed set, that ceiling doesn’t exist in the same way.
It adapts instead of just playing. For listeners using the heart-rate feature, the music responds to a measured physiological state rather than a manually chosen mood — the system maps the reading to an internal emotion scale and generates music to match. This is a feature of the personal listening experience; FeelFlow Radio doesn’t make any medical claims about it.
What this means if you’re evaluating it for a business
If you’re comparing this to how background music has traditionally worked in a shop, café, or restaurant, the practical difference is that there’s no separate licensing question to work through. Traditional background music — whether it’s a curated streaming playlist, a radio feed, or a hired DJ’s catalogue — involves someone else’s copyrighted recordings, which is why performing rights fees exist in the first place. Generated music doesn’t have that layer, because there’s no underlying recording to license.
Signing up follows the same logic: there’s no separate licensing negotiation, no consultant call, no contract to review clause by clause. You sign in, confirm your business and your location’s floor area, and the price is calculated automatically from that one number. For the fuller picture — what it costs, how the pricing works, and a direct comparison with what a shop typically pays today — see AI radio for business.