Put the arrangement and lyrics in separate fields

In the music generator, choose With vocals. Use Music prompt to describe the singer’s delivery, genre, mood, and instruments. Use Custom lyrics for the exact words you want sung. Choose a language that matches those words.

For an intimate acoustic song, a useful arrangement prompt is:

An intimate acoustic folk song with a clear solo vocal, fingerpicked nylon-string guitar, and soft upright bass. A relaxed 6/8 rhythm. Start with a brief guitar introduction and let the chorus open up slightly. Sing the supplied lyrics naturally.

Avoid asking for several conflicting vocal deliveries at once. “Quiet, close, and conversational” is more specific than a list of unrelated singer types.

Use verse and chorus labels

Google’s custom lyrics documentation recommends section tags such as [Verse], [Chorus], and [Bridge]. They describe song structure without becoming another paragraph of instructions.

This original example is also available in the A song for you template:

[Verse]
We took the long way through the town
And watched the evening settle down
No grand adventure, just us two
The ordinary felt brand new

[Chorus]
Here is a little song for you
For all the small things that you do
Wherever all our roads may go
You have a friend, I hope you know

Keep early experiments short. A verse and a chorus let you evaluate phrasing before investing in a longer song. PlayLyria sends custom lyrics as written; it does not automatically rewrite lines or add new sections to that field.

Make room for clear phrasing

Long lines may be rushed, stretched, or split across musical phrases. Read the lyrics aloud at approximately the tempo you want. If several words land awkwardly, shorten the line before generating another version.

For names or unusual words, clear spelling and a brief pronunciation cue in the music prompt may help, but the model is not guaranteed to follow it. Listen to the complete result before sharing it.

Understand what is and is not guaranteed

The model may repeat sections, change phrasing, or pronounce a word unexpectedly. Selecting a language is guidance, not a guarantee that every syllable will be sung correctly. This guide describes the input workflow; a tested vocal sample has not yet been published here.

Changing a prompt and generating again creates a new recording. It cannot replace just one line in your previous audio. Each new delivered track uses another credit.

Use lyrics you have permission to submit

Use your own writing or lyrics you are authorized to use. A model’s ability to sing text does not grant rights to that text. Review the Gemini API terms and any relevant third-party rights before distributing a song.