AI Song Generator: One Prompt, a Finished Song
Describe the track you hear, or bring the words you already have. Flow Music writes the arrangement, performs it, and hands back two takes plus a mixed master.
Most song generators return a loop and leave you to build the song around it. Flow Music starts from the other end: you write the brief for the finished track - mood, instrumentation, vocal direction, structure - and an arrangement comes back already mixed.
This page covers what belongs in a brief, what each studio control changes, what a render costs in credits, and what you can release once the two takes land.
How the AI song generator works
The generator is a single pass from a written brief to an arranged, performed and mixed track. Nothing is assembled from a sample pack and nothing is looped to length: the arrangement is written for the brief, performed, then mixed with the vocal sitting inside the instrumental rather than on top of it.
The four steps below are the whole workflow. There is no timeline, no plugin chain to build and no export dialogue to configure before you hear a result.
Write the brief
One paragraph is enough: the mood, the genre, who the song is for and the instruments you want to hear. Vocals and structure notes are optional at this stage.
Choose the shape
Set the length between 1:30 and 4:00 and decide whether the track carries a lead vocal, backing vocals only, or no vocal at all.
Generate and compare
Every render returns two takes with a producer log naming the arrangement decisions, so you hear what changed instead of guessing.
Keep, refine or export
Keep the take that works, ask for another pass with one changed detail, then download the master and, on paid plans, the separated stems.
What a good song prompt contains
A prompt that produces a usable song answers a handful of questions in order. Missing answers are not left blank - the model fills them with defaults, which is why a three-word prompt comes back sounding generic.
Thirty to sixty words is usually the sweet spot: long enough to cover mood, genre, instrumentation and vocal direction, short enough that no instruction gets diluted.
- Mood and story - what the song is about and how it should feel
- Genre and era - bedroom pop, synthwave, indie folk, drum and bass, and so on
- Instrumentation - the two or three instruments that must be audible
- Vocal direction - lead with harmonies, backing only, or instrumental
- Structure - where the chorus arrives, whether there is a bridge, how the track ends
- Things to avoid - no drums in the first verse, no vocal ad-libs, no fade-out
Every control in the studio
The studio exposes the decisions that change a render most, and keeps the rest automatic. Each control maps to something you can hear in the two takes that come back.
| Control | Options | What it changes |
|---|---|---|
| Mode | Prompt, Lyrics, Instrumental | Whether the studio writes the words, follows yours, or leaves the vocal out |
| Style | bedroom pop, synthwave, lo-fi, indie folk, afrobeat, cinematic, drum & bass, rnb | The palette, tempo range and production era of the arrangement |
| Length | 1:30 to 4:00 | How much arrangement fits in the render, and how many credits it uses |
| Vocal | Lead + harmonies, backing only, none | Whether a lead line carries the song or the track stays instrumental |
| Structure tags | [Verse], [Chorus], [Bridge], [Outro] | Where sections start when you supply your own lyrics |
| Takes | Two per render | Two arrangements of the same brief, so you can compare rather than accept |
What a render costs
Renders are metered in credits, and the estimate is shown before you confirm. Regenerating a take costs the same as the first render because it is a new arrangement, not a re-download.
| Render | Credits | When to use it |
|---|---|---|
| 1:30 clip | about 20 | Testing a brief before committing to a full song |
| Full song up to 4:00 | about 35 | The default render when you intend to release the track |
| Extended edit | about 60 | A longer arrangement with extra sections or a second chorus |
| Regenerate a take | about 35 | The same brief with one detail changed |
Prompt to render, in practice
A brief that reads "a rainy-night bedroom pop song about missing someone who moved away, warm pads, laid-back drums, soft female vocal, no drums in the first verse" gives the studio four instructions it can act on: the mood, the palette, the vocal and one structural rule.
The first pass typically returns a short intro, a verse that stays sparse, a chorus that opens up, and a second chorus that doubles the vocal. Take A usually keeps the dynamics small; Take B pushes the chorus wider and adds a counter-melody. That difference is the point of two takes - one of them will match what you had in your head, and you will know which within a few seconds.
From there, change one thing at a time. Swapping the style tag or removing the vocal tells you what that single decision was worth; rewriting the whole brief makes the result unattributable.
What you can do with the songs
Paid plans include commercial rights for the tracks you generate, so the master can go on streaming platforms, into client work or under a video. You do not have to clear samples because there are none: the arrangement is generated rather than assembled from a licensed library.
Keep your own records of what you wrote. Your lyrics remain yours, and the briefs you write are worth keeping with the export - they are the fastest way to produce a consistent follow-up track in the same session.
What the generator will not do
It will not recreate a named artist, a specific voice or an existing recording, and it will not master a recording you already own. It also will not fix a brief that contradicts itself: asking for a sparse ballad and a festival drop in the same prompt returns a compromise.
For work that needs a real performance - a vocalist you have booked, a band in a room - use the studio for the arrangement sketch and take the stems into the session rather than treating the render as final.
AI song generator questions, answered
What is the Flow Music AI song generator?
It is a browser studio that turns a written brief into a finished song: arrangement, performance, mix and a downloadable master, with separated stems on paid plans. You describe the track, or paste lyrics you already have, and two arrangements come back per render.
Do I need to know music theory?
No. The controls are musical rather than technical: mode, style, length, vocal presence and, if you bring lyrics, structure tags. If you can describe the song you want to hear, you can direct the studio.
How long does a song take to generate?
A full render is a couple of minutes. The producer log shows each stage - arrangement, topline, tracking, mix - so you can see where the render is instead of watching a spinner.
What happens if I do not like either take?
Change one thing in the brief and regenerate. Two new takes come back with the same two-arrangement structure, and you keep every version you generated until you decide which one to export.
Can I generate a song from my own lyrics?
Yes. Switch to lyrics mode, paste up to 2,000 characters and mark sections with tags like [Verse] and [Chorus]. The studio writes the music around your words instead of writing new ones.
Can I make an instrumental instead?
Choose Instrumental only and the vocal is left out entirely. That is what most video beds, podcast intros and game cues need, and it costs the same as a vocal render of the same length.
How many credits does a song use?
A 1:30 clip is about 20 credits, a full song about 35 and an extended edit about 60. Regenerating a take costs the same as the render that produced it.
Can I release the songs I generate?
Yes. Paid plans include commercial rights for the tracks you generate. You can publish them on streaming platforms, use them in client work, or put them under video - there are no samples to clear.