Guide to Prompting for Music

In the Create music Action, your prompt is your description of the song. A short prompt already works, because the model fills in the rest, so detail is a way to take control rather than a requirement.

You have up to 4,056 characters, and length and number of variations are set below the prompt box, not written in the prompt. Music is generated with Google Lyria 3 Pro.

Writing in any language

You can prompt in languages other than English, and you can mix them in one prompt. Keep the musical instructions (tempo, genre, voice) in English and paste your lyrics in another language, write the whole prompt in another language, or any combination. Use whichever language is most natural for the part you are describing.

Genre, style, and sound

Name the genre and style, and feel free to combine more than one. You can anchor a sound to an era and get specific about the building blocks:

• Genre and style, or a blend of two: "indie folk", "synthwave meets trap"
• A year or era: "80s", "early 2000s R&B"
• Tempo in BPM: "124 BPM", or in words: "slow and unhurried"
• Key or scale: "in C minor", "F-sharp major"
• Instruments to feature: "fingerpicked acoustic guitar, soft brushed drums"
• Mood and atmosphere: "wistful", "menacing", "shimmering", "hypnotic".

If you want an instrumental piece, state it explicitly so the model leaves out vocals.

Song structure

You can outline how the song progresses. Use a list or arrows to map the flow:

• [Soft intro] -> [Verse 1] -> [Pre-chorus] -> [Chorus] -> [Verse 2] -> [Chorus] -> [Bridge] -> [Final chorus]
• Open on a lone acoustic guitar, add drums for the first verse, then pull everything back just before the chorus lands at full volume.

You can describe how the energy shifts from section to section:

• Let the second verse run hot, then cut to almost nothing for a half-time bridge
• Fill the arrangement out a little more every eight bars until it feels huge
• Hold the last chorus back, then bring it in louder than anything before it.

You can also call out the exact moment you want something to happen:

• Land the first big drop around the 15-second mark
• A handclap on every other beat
• Bring the full band in at 0:20.

You can prefix parts of the song with section titles like [Intro], [Verse 1], [Pre-chorus], [Chorus], and [Outro]. To repeat a word or phrase, like a vocal echo or a backing-vocal answer, wrap it in parentheses:

• Hold on (hold on).

Lyrics

If you want the model to write the lyrics, tell it what they should be about. Without a subject it has to guess one from the music description, and it may not land where you wanted.

• Write lyrics about leaving a small town at dawn and not looking back, with a restless, hopeful feeling.

If you want a chorus that comes back, ask for one:

• ...and give it a big, repeatable chorus about starting over.

The model already shapes the lyrics to fit the style you ask for, but you can reinforce it:

• A dance track built around a single hook that keeps returning.

You can also ask for vocal moments that are not really lyrics:

• A crackly radio voice introduces the track before the beat kicks in
• Right before the final drop everything cuts out and a whispered voice counts to three
• The song opens mid-conversation about an old road trip, then slides into a warm chorus.

Vocals

You can describe how the vocals are delivered. A detailed singer profile, covering gender, tone, and range, gives the most reliable results:

• A bright female voice with a clear, weightless top end and an easy, soaring quality
• A warm, husky female alto with a smoky edge and a soulful, lived-in tone
• An energetic male tenor, slightly raspy, that cuts straight through a loud mix
• A deep, smooth male baritone with a calm, crooning delivery
• A worn, gravelly rock voice that strains at the top for extra intensity.

Putting it together

Start small and add detail only where you want control. A few examples, from minimal to detailed:

• a slow piano ballad about missing home
• 110 BPM dream-pop in A minor, hazy and bittersweet, instrumental, building one layer at a time
• an anthemic rock song about second chances with a huge repeating chorus, sung by a worn, gravelly male voice, moving from a quiet intro through two verses into a final chorus that doubles in size.