AI Music Generator: A Style Prompt In, a Finished Song Out

A music generator is only useful when the track comes back finished. VdoBloom's version starts from two text boxes: a style description that is required, and lyrics you paste yourself that are optional. The output is a downloadable audio file — an instrumental bed or a sung song with a vocal in the language you name — with no arrangement, mixing or session-musician step in between.

Three models sit behind the same form. MiniMax Music 2.6 is the full-song option, taking your lyrics and singing them, with an Instrumental switch that drops the vocal and returns a backing track instead. ACE-Step Base and ACE-Step Turbo work differently: you set a duration anywhere from 30 to 300 seconds, name the vocal language, and let the model build the track to that length. The style prompt is the one thing every model needs, and it is also the thing that decides whether the result sounds like the song in your head.

Generate a song

What this tool does

  • VdoBloom's AI music generator offers three models: MiniMax Music 2.6 for full songs with lyrics and vocals, plus ACE-Step Base and ACE-Step Turbo.
  • The style/description prompt is required; the lyrics box is optional, and MiniMax Music 2.6 has an Instrumental switch that returns a backing track with no vocal.
  • The ACE-Step models take a duration you set between 30 and 300 seconds and a vocal-language field, so a track can run from half a minute to five minutes.
  • The words are yours: the tool sets lyrics you supply to music, and there is no automatic lyric writer in the form.
  • Every finished track is stored in your audio library with its prompt, model and lyrics, so it can be played or downloaded again later.
  • New accounts start with free signup credits, one-time credit packs never expire, and subscriptions are also available for a monthly allowance.

What makes a style prompt work

The style box is not a search field, and it is not a place for a sentence about how the song should make someone feel. It is a description of a recording: which genre, which instruments carry it, how the drums sit, how bright or dark the tone is, how fast it moves. "Slow acoustic folk, fingerpicked guitar, brushed snare, warm upright bass, intimate close vocal" will come back much closer to that description than "emotional ballad".

Tempo is worth naming in words even when there is no BPM box, because the model reads "half-time", "driving four-on-the-floor", "slow triplet feel" and "mid-tempo shuffle" as real instructions. If the first pass comes back too busy, the fix is usually to subtract: fewer instruments, one clear lead, and a single rhythmic idea rather than three.

Lyrics, verses and the second pass

When you paste lyrics, structure them the way a songwriter would. Keep one line per line, mark the sections you want — verse, chorus, bridge — and keep the chorus shorter than the verses so the repetition reads as a hook rather than padding. Models handle a short, repeated, metrically regular chorus far more reliably than a long free-form block of text.

Generated songs rarely land perfectly the first time, and the honest workflow is iterative: keep the style prompt that sounded closest, change one thing — the duration, one instrument, the chorus wording — and regenerate. Because each run is independent, the version that works is often the third or fourth, so listen before you commit to using a take rather than judging from the first few seconds.

How it works

  1. 1.Write the style description

    Say what you want to hear — genre, instruments, mood, tempo feel. "Warm lo-fi piano with vinyl crackle and soft brushed drums" gives the model far more to work with than "sad song".

  2. 2.Paste lyrics, or leave them out

    Paste the words you want sung, line by line. Tick Instrumental on MiniMax Music 2.6 to drop the vocal entirely; on the ACE-Step models, simply leave the lyrics box empty.

  3. 3.Choose a model, length and language

    MiniMax Music 2.6 for a complete song with vocals. ACE-Step Base or Turbo if you want to fix the duration between 30 and 300 seconds, and to name the language the vocal is sung in.

  4. 4.Generate, then listen end to end

    The finished track lands in your audio library. Listen all the way through before you use it — intros, outros and the last chorus are where generated songs most often drift from the brief.

Frequently asked questions

Does the AI write the lyrics for me?

No. The lyrics box takes words you supply, and if you leave it empty you get no vocal on the ACE-Step models, or an instrumental on MiniMax Music 2.6 if you tick the switch. There is no lyric-writing step in this tool, which is also why the copyright position is simple: the words in the song are the words you typed.

What kinds of music can it make?

The style prompt is free text, so it can describe anything: pop, rock, EDM, acoustic singer-songwriter, orchestral, hip-hop, lo-fi study beats, cinematic score, devotional music. Two or three concrete details work better than one adjective — name a genre, an instrument and a mood, and the result lands much closer to what you meant.

How long can a generated song be?

With ACE-Step Base or ACE-Step Turbo you set the duration directly, anywhere from 30 to 300 seconds. With MiniMax Music 2.6 the length follows the song structure the model returns for your lyrics, so it is not something you dial in on the form.

Can I generate an instrumental with no vocals?

Yes. On MiniMax Music 2.6 tick Instrumental and the lyrics field disappears from the form, which returns a backing track from the same style prompt. On either ACE-Step model, leave the lyrics box empty and the track comes back without a sung line.

Can the vocal be in another language?

The ACE-Step models expose a vocal-language field, so you name the language you want the singer in. MiniMax Music 2.6 follows the lyrics you paste, so writing the words in your own language is how you steer it. Either way, proof-listen the vocal before you publish — pronunciation in less common languages is the part most worth checking.

Is it free to try, and do I need a subscription?

New accounts start with free signup credits, so you can hear what the models do before deciding anything. After that, generation runs on credits: credits bought in one-time packs never expire, and subscriptions exist for anyone who would rather have a monthly allowance.

Can I use a generated song in a video, a podcast or a paid ad?

Under VdoBloom's terms you keep the content you upload or generate and you are responsible for it. That covers putting the track under a video, a podcast intro or a social post. Two things to keep clean: the lyrics must be yours or properly licensed, and if you are writing for a client, check the terms of the engagement before you hand the file over.

Related tools

AI Rap Song Generator: Your Bars Over a Produced Beat

Write or paste your bars, describe the beat and get a finished rap track back — vocals over a produced instrumental. Free signup credits for new accounts.

AI Lo-Fi Music Generator: Dusty Beats, Made to Order

Describe the lo-fi sound you want — dusty piano, vinyl crackle, soft drums — and get a finished track back, instrumental or with vocals, no studio needed.

AI Song Cover Generator: Your Song, a Different Record

Upload a song you own and hear it in a new genre, extend it, add vocals to your instrumental or a backing track under your voice. Rights confirmed.

AI Lyric Video Maker: Every Beat Gets a Shot

Turn a song into a beat-synced video with the lyrics on screen — generate the track or upload your own, lock one character, and cut every shot on the beat.

Ready to try it?

Generate a song