Make your first song

For anyone who just signed up and is looking at an empty prompt box.5 min read

You do not need to understand the settings to get a song out of this. Pick one of the two ways in, spend one generation, then come back and read what the controls do — they will make far more sense once you have something to compare against.

01

Pick the way in that matches what you have

The editor opens on two tabs, Simple and Advanced, and the only question that decides between them is whether you already have words.

SimpleAdvanced
What you bringOne sentence describing the songYour own lyrics, plus a style description
What you controlMood, genre, whether it is instrumentalLyrics, section structure, style, vocals, reference audio
Best forHearing what the model does with an ideaA song you already have in your head

02

Your first generation, step by step

  1. 1

    Describe the song in one sentence

    Say the mood, the genre and the voice — for example, a warm folk song for my mother, acoustic guitar and a soft female vocal. If nothing comes to mind, use one of the ready-made ideas under the box; they are written to show the shape a good description takes.

  2. 2

    Generate, then listen all the way through

    Resist judging it in the first ten seconds. The intro is the least characteristic part of any generated track — the chorus is where you find out whether the model understood you.

  3. 3

    Change exactly one thing and run it again

    One variable at a time is the whole discipline. Swap the genre, or the vocal, or the mood — not all three. Two generations that differ in one word teach you more than ten that differ in everything.

03

What the controls actually change

Once you have a track to compare against, these are worth turning on. Every one of them exists to remove a guess.

  • Enhance Prompt rewrites a rough style note into a fuller description before you spend anything on it. Read what it produces — it is the fastest way to learn the vocabulary.
  • Vocals sets male, female or any, instead of hoping the model picks the one you meant.
  • Exclude styles is the negative prompt. Use it for the things that keep showing up uninvited, not for things you simply are not asking for.
  • Add Reference Audio steers the result toward a track you already like, on the Advanced tab.
  • Compare Models sends the same prompt to several models at once, so you find out whether the prompt or the model was the problem.

04

After the song exists

Everything below happens on a finished track, from your Library.

  • Regenerate Artwork replaces the cover from a text description of what you want.
  • Create a Lyric Video syncs the lyrics to the audio and exports a shareable video.
  • Stem Separation splits the track into vocals and instruments, and gives you the stems as lossless files.

Common questions

How many songs do I get before paying?

You start with credits on the account, and free generations are not watermarked, not shortened, and not run on a weaker model than paid members use. What a paid membership adds is commercial use and a larger credit budget.

My first song sounds nothing like what I asked for. What now?

That is the normal first result, and it is usually a vocabulary gap rather than a model failure. Run the same idea through Compare Models to see whether another model reads it differently, or analyse a reference track to get language that describes the sound you are hearing in your head.

Where do my songs go?

Every generation is saved to your Library automatically, including the ones you do not like. Nothing is discarded unless you move it to the trash yourself.

Spend one generation now

Reading about prompting teaches you very little. One sentence, one song, then come back to this page.

Open the editor

Read next