Upload a demo, describe the genre you want to hear, and get the same song back with a new voice, new instruments and a new feel. Try your ballad as jazz, your acoustic idea as hard rock or your pop hook as a lo-fi loop before you book a single studio hour.
A demo is a song in its first clothes. The melody works and the words are there, but the arrangement is whatever you could record on a quiet evening with the gear on your desk. Restyling lets you hear that same song as a jazz ballad, a rock track or a lo-fi loop before you commit to a single studio hour. Upload the demo, describe the genre, and compare what comes back with the version in your head.
The usual way to find out whether a song works in another genre is to hire players or lose a weekend building a second arrangement in a recording program. That is a lot of effort for a hunch. Restyling turns the hunch into a listening test: minimax/music-cover takes your MP3 or WAV file, keeps the melody, and rebuilds the voice, the instruments and the feel from a text description.
It is most useful at the decision stage. A songwriter who suspects a folk tune wants drums, a band weighing a slower version for the ballad slot, a podcaster who needs the same theme in a calmer mood: all of them are asking the same question, and a few generated versions answer it faster than a rehearsal does. The model is one of ten in the music generation category, out of 502 on Picasso IA, and you can browse its neighbors in the full catalog or by category. If you do not have a song yet and only have words, song lyrics to music is the other direction.
Genre words do most of the work in a restyle prompt, but instrument and vocal words do the rest. A genre name alone leaves the model to guess which version of the genre you mean, so pair it with a lead instrument, a drum feel and a voice type. The table below uses the kind of phrasing the model page itself shows in its examples, plus what to listen for when the result comes back.
| Target style | Prompt words that steer it | Listen for | Fits best |
|---|---|---|---|
| Jazz | saxophone lead, mellow piano chords, brushed drums, late night club | Whether the melody still reads over the richer chords | Slow songs and ballads |
| Rock | distorted electric guitar riffs, powerful drums, raw vocal, arena energy | Whether the vocal stays clear above the guitars | Mid-tempo songs with a strong hook |
| Lo-fi | vinyl crackle, dusty drums, warm midrange, dreamy reverb | Whether the lyrics still land at low energy | Simple melodies and chill playlists |
| Acoustic folk | fingerpicked guitar, soft vocal, no drums | How much the song leaned on its beat | Pop or hip-hop with a clear tune |
| Bossa nova | nylon string guitar, soft brushed drums, warm vocal | Whether the rhythm sits naturally under the phrasing | Gentle, melodic songs |
Run the same demo through two or three rows rather than one. The comparison is the point: a song that falls apart as rock may come alive as bossa nova, and you only find out by hearing both. If the lo-fi version wins, the same sound is what people look for in lofi beats for streams and study, so that page is a useful reference for the genre vocabulary.
This is the shortest path from a rough recording to a set of genre versions you can compare side by side. It assumes a finished demo file and about twenty minutes of listening time.
The model keeps your melody on its own, so the prompt should describe everything around it. Think of it as a brief to a session band: what instruments are in the room, how hard the drummer plays, what the singer sounds like, how big the space is. Words about tempo feel and production, such as laid back, driving or dry and close, steer the result as much as the genre name does.
Tip: change the genre, not the song. Describe instruments, rhythm and vocal color, and leave the melody out of the prompt. The model already carries it over, and asking for a different tune only pulls the result away from the demo you wanted to hear in a new style.
Keep a short log as you go: the prompt, the genre, one line about what you liked. After four or five runs the pattern shows up on its own, for example that your chorus needs a warm bass and your verse needs space. The model also lets you pick the sample rate and bitrate of the output, so use the smaller options while you are only auditioning and save the highest quality for the version you keep.
A restyle is an audition, not a finished record. These are the places it stops helping:
When a version is good enough to build on, take it to your players or your producer as a reference recording. That is where it saves real money: the conversation about the arrangement starts from something they can hear.
Use this on songs you wrote, songs you co-own with the people who agreed to it, and tracks whose license allows derivative versions. A cover of someone else's song is still built on their work, and switching genre does not make it yours. The same applies to vocals: describe a type of voice, never a named artist.
If the result is headed for streaming, check the current requirements of the platform and your distributor before you upload, since rules on AI-assisted music and on cover versions differ and change. This page is general information, not legal advice; for anything commercial, a music lawyer or your distributor can tell you what applies where you live. Once the track is ready, album cover art can be made in the same toolkit.
Yes. On Picasso IA the minimax/music-cover model takes an MP3 or WAV of your song and a text description of the target style, then returns the song rebuilt in that style. The melody carries through while the voice, the instruments and the genre change. Jazz, rock and lo-fi are the easiest to describe, but any genre you can put into words is worth a try, from folk to bossa nova to dance remixes.
minimax/music-cover is the one built for it. It is a cover model: you give it your audio and a style prompt, and it rebuilds the song. The other music models in the catalog, such as minimax/music-2.6, google/lyria-3 and elevenlabs/music, are generators that create new tracks from a description, which is a different job. Check the model page in the catalog for the current inputs before you start.
The melody is the part the model is designed to keep. The lyrics stay too unless you replace them: if you leave the lyrics field empty, the singer performs the words found in your source song, and if you paste new lyrics they override the originals. Results vary from run to run, so listen to the whole track rather than the first ten seconds, and regenerate if a phrase drifts.
An MP3 or a WAV of the song you want to restyle. A clean export of your demo works best: clear vocal, no clipping, no long silence at the start. If the source is a rough phone recording, expect the result to inherit some of that roughness in the melody it follows. Output can come back as MP3, WAV or PCM, so pick the format your next step needs.
Yes. The lyrics field accepts new words, up to 3000 characters, and you can mark sections with tags such as [verse], [chorus] and [bridge] so the singer follows the structure. This is handy for trying an alternate hook or a translated verse. If you leave it empty, the model keeps the lyrics from your original song, so you only fill it in when you want a change.
Only if you have the rights to do so. Restyling a track you do not own, or turning a famous song into a new genre and releasing it, can infringe the original writer and recording owner even though the sound is different. Stick to your own compositions and tracks whose license clearly allows derivative versions. This is general information, not legal advice, and a music lawyer can answer your specific case.
Neither. The model builds a voice from your prompt, such as smooth female vocal or raw male vocal, and it will not reproduce your own voice. Do not try to imitate a named artist, since using a real person's voice without permission is a legal and ethical problem. If your own voice matters to the song, treat the restyle as a reference and record the final vocal yourself.
Treat it as an audition. You get one audio file back, not separate stems, and the arrangement was generated rather than mixed by an engineer. Many people use the result as a reference to brief musicians or to decide which genre to pursue. If you do release something built on it, check the current requirements of your distributor and streaming platform on AI-assisted music first.
Be specific about instruments, drum feel and vocal character, and change one phrase per run so you can tell what moved the result. Compare two genres from the same demo instead of polishing one prompt forever. Start with a cleaner source file if the melody keeps drifting. Words like warm, dry, laid back or driving help as much as the genre name does.
Generation costs credits, and the amount depends on the model and the settings you choose, so any number written here would go stale. New accounts get free credits, which is enough to try a few genres on one demo. Current plans are listed on the pricing page, the only place worth trusting for numbers. If you have questions about a specific run, the support page is the place to ask.
Your demo already has a melody worth hearing in another genre. Open the Picasso IA toolkit, pick the cover model, and listen to your own song in jazz first.
Choose your language
