The prompt formula behind good AI music
An AI music model reads your prompt as a list of musical facts. The more of the right facts it gets, the less it has to guess. A formula that works across genres has five parts, and you can write it in a single line.
- Genre and sub-genre. Not just rock, but indie rock or garage punk. This sets the drums, the guitar tone and the song structure more than anything else.
- Vocal. Say bright female vocal, gentle male vocal, male rap vocal, choir, or instrumental. If you say nothing, the model chooses for you.
- Two or three instruments. Fingerpicked guitar and soft cello, or Rhodes piano and vinyl crackle. Four or more instruments tends to blur into one wash.
- Mood and energy. Nostalgic and tender, confident and late night, euphoric build. Add a tempo such as 112 bpm if you care about it.
- A scene. A road trip along the coast, a rainy bus ride home. Scenes give the model a story to follow.
Here is the formula applied. Upbeat indie pop, bright female vocal, warm synths and claps, summer road trip feeling, 112 bpm. That is a real prompt used on this site, and the next sections show what it produced. If you have lyrics, paste them in the lyrics tab and keep the prompt for the sound only. The prompt then describes how the song should sound, and the lyrics decide what it says.
The one thing to avoid is naming a real artist. Prompts that ask for a named singer are against the rules here, and a good prompt does not need them. Describe the sound itself, which is also more reliable.
Genre by genre: prompts and what to expect
Use this table as a starting point. The prompts are written to be copied and edited. The expectations column describes what the style typically gives you from this kind of model, not a promise, because every generation is different. The rows for indie pop, boom bap, folk and lo-fi are based on real generations listed later on this page. The rest are common styles from the style list in the generator.
| Genre | Prompt to try | What to expect |
|---|---|---|
| Pop | Bright modern pop, female vocal, punchy drums, synth bass, big chorus, 118 bpm | A clear verse and chorus shape, a hook, polished vocal. The safest genre to start with. |
| Indie pop | Upbeat indie pop, bright female vocal, warm synths, claps, summer road trip feeling, 112 bpm | Light, airy and catchy. Works well with short, rhymed lyrics. |
| Lo-fi hip hop | lofi hip hop, chill, rhodes piano, vinyl crackle, instrumental | Loopable, mellow and fast to generate. Ideal for study or stream background. |
| Boom bap hip hop | Boom bap hip hop, confident male rap vocal, dusty vinyl drums, jazzy piano loop, late night city | Steady swung drums and a rhythmic vocal. Lyrics with strong internal rhyme help. |
| Trap | Dark trap, 140 bpm, rolling hi-hats, heavy 808 bass, male vocal, moody | Hard low end and fast hats. Write short, repetitive hooks. |
| Acoustic folk | Acoustic folk ballad, gentle male vocal, fingerpicked guitar, soft cello, nostalgic and tender | Intimate, slower, story-driven. Narrative lyrics suit it well. |
| Rock | Driving indie rock, male vocal, overdriven guitars, live drums, anthemic chorus, 128 bpm | Guitar-led with a loud chorus. Give it a clear section structure. |
| R&B | Smooth R&B, female vocal, warm bass, finger snaps, late night, 82 bpm | Relaxed groove and layered vocal. Keep lyrics simple and melodic. |
| EDM | Euphoric progressive house, 124 bpm, big build and drop, plucky synths, instrumental | Builds and drops. Works best as an instrumental. |
| Country | Modern country, male vocal, acoustic guitar, pedal steel, storytelling, mid tempo | Warm and narrative. Specific details in the lyrics help. |
| Jazz | Cool jazz trio, upright bass, brushed drums, piano, relaxed late night, instrumental | Small-band feel. Instrumental prompts tend to behave best here. |
| Cinematic | Epic cinematic orchestral, strings and brass, taiko drums, slow build, no vocals | Slow swells and big climaxes. Good for trailers and game scenes. |
| K-pop | Energetic K-pop, group vocals, bright synths, tight dance beat, catchy hook, 125 bpm | High energy and layered vocals. Short lines repeat better than long ones. |
| Reggaeton | Reggaeton, dembow rhythm, male vocal, warm bass, summer party, 95 bpm | Distinct dembow groove. Spanish lyrics work if you paste them in. |
If a genre sounds close but not right, change the tempo or swap one instrument before rewriting the whole prompt. Small edits teach you what the model responds to, and you only get three generations a day on the free plan.
Hear real prompts and results
These are unedited generations made while building the site, with the exact prompts listed. Notice how the same indie pop prompt gives two different arrangements, one from each model. The lo-fi track is an instrumental and arrived far faster than the vocal songs.
Sunlight A (MiniMax Music v2, 2:30) took 94 seconds. Sunlight B (ACE-Step, 1:44) took 27 seconds. Lo-fi Study (ACE-Step, 2:00, instrumental) took 11 seconds. Late Shift (MiniMax Music v2, 1:08) took 59 seconds. Times were measured from request to finished file, and your own times will vary with load.
Instrumental or vocal: which to choose
The vocal switch is the biggest single choice after genre, and the right answer depends on what the track is for. The generator lets you pick female, male or none, and none produces an instrumental.
| Need | Choose | Why |
|---|---|---|
| Background music for a video, stream or podcast | Instrumental | No lyrics to compete with speech, and a loopable bed is easier to edit. |
| Study, focus or sleep playlist | Instrumental | Lo-fi, ambient and piano prompts are quick and consistent. |
| A gift, birthday or personal message | Vocal with your lyrics | The words are the point. Paste lyrics and keep the prompt for sound. |
| A demo of a song idea | Vocal | Gives you a melody and a feel to react to and refine. |
| Game, trailer or cinematic scene | Instrumental | Orchestral and electronic scores sit under dialogue and effects. |
| Rap or sung hook | Vocal | Rhythm of the delivery is part of the genre. |
A useful trick is to make a vocal song you like, then split it. The stem splitter on this site separates vocals, drums, bass and other, so you can keep the instrumental or the acapella of a track you made. Only upload audio you have the rights to.
Two models, two takes: MiniMax Music v2 and ACE-Step
Every generation runs two models and gives you both results. They behave differently, which is why one prompt yields two distinct songs rather than a duplicate. The figures below come from the timings listed earlier on this page, not from a benchmark.
| ACE-Step | MiniMax Music v2 | |
|---|---|---|
| Typical speed in our tests | 11 to 27 seconds | 59 to 94 seconds |
| Best use | Quick sketches, instrumentals, checking the direction | Fuller vocal songs where you want a richer take |
| Longest result seen | 2:00 (instrumental) | 2:30 |
| Cost on this site | Included in the 10 credits | Included in the 10 credits |
In practice, listen to the quick ACE-Step version first. If the genre and tempo are right, wait for the MiniMax version and compare. If the quick version is wrong, you know to change the prompt without waiting. Neither model is always better, so keeping both in the results is the honest way to use them.
Rights: what you can do with the music
Licensing is where AI music tools differ most, so read this part before you publish anything. These points follow the TuneSplit Terms of Use.
- Songs made on the free tier are for personal, non-commercial use: sharing with friends, personal posts without monetisation, practice.
- Songs made with an active paid plan or purchased credits may be used commercially, including monetised YouTube, podcasts, ads, games, apps and streaming distribution. The licence for those songs stays valid after you cancel.
- AI output can resemble existing music by chance. You are responsible for checking that your use does not infringe anyone's rights, and for any lyrics you write or paste.
- There is no voice cloning, no AI covers and no prompts asking for a real, named singer or artist.
Paid plans are Pro $9.90 a month or $99 a year, and Premier $24.90 a month or $259 a year. Credit packs are 200 credits for $4.90 and 600 for $12.90. The pricing page always has the current numbers.
Free limits and what happens to your prompts
The free plan is simple to state. There is no sign-up. Each browser gets 30 credits every UTC day, reset at 00:00 UTC. A generation costs 10 credits and returns two songs, so that is 3 generations and 6 songs, each up to 2 minutes, with MP3 download. Unused credits do not roll over. If a generation fails, its credits are returned automatically.
On privacy, the generator does not take audio uploads, so there is no upload to delete. What the site stores is your prompt or lyrics, the style settings and the generated audio, so you can play and download them later. A song is public only if you share it or submit it to the Daily Song, and you can make it private again from My songs. Prompts and lyrics are sent to fal.ai, which runs the models, only to produce your result. Analytics record that a song was generated, without your prompts or lyrics. To delete a song, email the address on the privacy page.
AI Music Generator facts at a glance
| Fact | Detail |
|---|---|
| Price | Free tier: first song free, then 30 credits a day (3 generations, 6 songs). Paid plans from $9.90 a month |
| Download format | MP3 |
| Sign-up | Not needed. Credits are counted per browser |
| File deletion | No audio is uploaded to this tool. Prompts, lyrics and songs are kept until you ask for deletion, and you can make a song private at any time |
| Models | MiniMax Music v2 and ACE-Step, both on every generation |
| Speed | ACE-Step about 10 to 30 seconds, MiniMax about 60 to 95 seconds in our tests |
| Limits | Songs up to 2 minutes on free, 3 on Pro and 4 on Premier. Free songs are personal use only |
Get better results: a quick revision loop
Treat the free daily generations as three rounds of a small experiment rather than three lottery tickets.
- Round one: write the five-part formula and generate. Judge only the genre and tempo.
- Round two: change a single variable, for example swap pedal steel for piano or lower the tempo by 15 bpm.
- Round three: add your lyrics, or tighten them, and keep the winning sound prompt.
- Download the best MP3. If you want it longer or commercial, that is where a paid plan helps.
If you need lyrics first, the AI lyrics generator on this site drafts them for free, five a day, and sends them to the song generator in one click.
Frequently asked questions
Is this AI music generator really free?
Yes. There is no sign-up and no card. The free plan gives each browser 30 credits every UTC day. One generation costs 10 credits and returns two versions, so you get 3 generations and 6 songs a day, each up to 2 minutes, with MP3 download. Free songs are for personal use.
Do I need to sign up to make music?
No. You can write a prompt, generate and download straight away. The daily 30 credits are counted per browser, and your songs are saved in this browser under My songs. The allowance resets at 00:00 UTC each day and unused credits do not carry over to the next one.
Can I download the music I generate?
Yes. Every song has an MP3 download on the free plan, and every song also gets a share page. Some AI music tools restrict downloads on their free plans, so it is worth checking. If you want separate vocal and instrument tracks of your own song, the full-length stems are a paid feature.
What is the best prompt for an AI music generator?
Use five parts: genre and sub-genre, vocal type, two or three instruments, mood with an optional tempo, and a short scene. For example, boom bap hip hop, confident male rap vocal, dusty drums, jazzy piano loop, late night city. Avoid naming real artists, and change one detail at a time when you iterate.
Can it make instrumental music with no vocals?
Yes. Choose no vocal in the Advanced section, or write instrumental in your prompt. Lo-fi, ambient, piano, cinematic and house prompts work well this way. A lo-fi instrumental in our tests took about 11 seconds from the ACE-Step model, so instrumentals are also the quickest way to experiment.
Can I use my lyrics?
Yes. Open the Advanced section, switch to the lyrics tab and paste your words. Keep the prompt for the sound only, for example genre, vocal and instruments. Lyrics are checked automatically, and a request can be refused if it breaks the rules, such as harassment, hate or copying someone else's lyrics.
How are the two versions different?
Each generation runs two models. ACE-Step usually returns in 10 to 30 seconds and is good for fast sketches. MiniMax Music v2 usually takes 60 to 95 seconds and gives a fuller take. They interpret the same prompt differently, so you get two arrangements to compare, and you can keep either one.
How long can the songs be?
Free songs run up to 2 minutes. The Pro plan raises that to 3 minutes and Premier to 4 minutes. Actual length depends on the model and how much lyric text you provide. In our tests, results ranged from about 1 minute 8 seconds to 2 minutes 30 seconds.
Can I use AI-generated songs commercially?
Only with a paid plan or purchased credits. Songs made on the free tier are for personal, non-commercial use. Songs made while you have an active paid plan or with bought credits may be used commercially, including monetised YouTube, podcasts, ads and streaming, and that licence stays valid after you cancel.
Can it copy a famous singer or clone a voice?
No. Voice cloning, AI covers and prompts that ask for a real, named singer or artist are not allowed. Describe the sound instead, for example warm female alto vocal with a breathy tone. This is a rule of the site, and it also tends to give you more original results.
What happens to my prompts and songs?
Your prompt or lyrics, style settings and generated audio are stored so you can play and download them later. A song is public only if you share it or enter it in the Daily Song, and you can make it private again. Prompts go to fal.ai, which runs the models, only to make your result.
What if a generation fails or sounds wrong?
If a generation fails, its 10 credits are refunded automatically. If it simply sounds wrong, adjust one variable such as genre, tempo or one instrument and run it again. The table of genre prompts on this page is a good place to start, and listening to both versions often shows which part to change.