Video tutorial · 1:27

How to write AI song prompts without naming an artist

AI song prompts work best when they describe the sound, not the singer. About 1 in 10 prompts in our study of 6,798 AI songs tried to copy a real artist, and most song models are built to refuse that. Name four ingredients instead: genre, mood, instruments and voice.

Make a song →Read the transcript

Key takeaways

Key takeaways for AI song prompts

A song model listens for ingredients, not names. Each of the four below tells the model something it can actually use, so you get fewer refusals and fewer bland takes. Keep the list next to you the next time you write a prompt.

01

Genre

The style family, like indie folk or synthwave.

02

Mood

How it should feel, like warm and a little sad.

03

Instruments

What plays, like fingerpicked guitar and soft piano.

04

Voice

How the vocals sound, like soft and breathy, close to the mic.

Example

AI song prompt example: before and after

Both prompts aim for the same song. The first one names a person, which most models won't imitate, so they refuse or guess. The second one describes the genre, mood, instruments and voice, so the model knows exactly what to make.

Before · a name

“make it sound like my favorite singer”

After · a recipe

“Melancholic indie folk, fingerpicked guitar, soft breathy vocals, slow and warm.”

Transcript

Video transcript

The full narration, split into the seven chapters of the video. Select a timestamp to play the video from that moment, or read straight through if you prefer text.

Why AI refuses your prompt

You asked an AI to make a song that sounds like your favorite singer. And it refused, or gave you something flat.

Naming people

We all describe music by naming people. Make it sound like that famous voice. That is how we talk to friends.

What the data shows

In our study of almost 7,000 AI songs, about one in ten prompts tried to copy a real artist. But most song models are built to avoid copying real artists. So they refuse, or they guess, and the guess comes out bland.

The 4 ingredients

A song model listens for ingredients, not names. Give it four. A genre, like indie folk. A mood, like warm and a little sad. Instruments, like fingerpicked guitar and soft piano. And a voice, like soft, breathy vocals close to the mic.

Same idea, done right

Now try again. Melancholic indie folk, fingerpicked guitar, soft breathy vocals, slow and warm. No names at all. And this time, the model knows exactly what to make.

Why it works better

Fewer refusals, and fewer bland takes. The song sounds like your idea, not a weak copy of someone else. And you can reuse that recipe for every song you make.

Recap

Before you hit create, check four things. Genre. Mood. Instruments. Voice. Describe the sound, not the singer. Make your next song at onmusic.ai.

AI song prompts FAQ

Why do AI song generators refuse prompts with artist names?

Most AI song models are built to avoid imitating specific real artists, so prompts that name a singer are often rejected or rewritten. In our study of 6,798 AI songs, about 1 in 10 prompts tried to copy a real artist. Describing the sound instead avoids the problem.

What should AI song prompts include?

AI song prompts should name four ingredients: a genre, a mood, the instruments and the voice. For example, melancholic indie folk, fingerpicked guitar, soft breathy vocals, slow and warm gives the model everything it needs without naming anyone.

How do I describe a singer's voice without naming them?

Describe how the voice sounds rather than who it belongs to: soft or powerful, breathy or clear, low or high, close to the mic or distant. Add the delivery too, like whispered verses or a belted chorus.

Can I reuse the same AI song prompts for more songs?

Once a genre, mood, instruments and voice combination works, save those AI song prompts as a recipe. Change one ingredient at a time, like the mood or the tempo, to make a matching set of songs that still sound like your idea.

How do I choose a genre for AI song prompts?

Pick the closest genre plus one modifier, like dreamy synth-pop or gritty blues rock. A specific pair tells the model more than a broad label like pop, and it keeps the arrangement consistent from verse to chorus.

Should AI song prompts include a tempo?

A rough tempo helps. Words like slow, mid-tempo or upbeat work, and so do numbers such as 90 BPM. Tempo shapes the energy of the whole track, so add it when the mood alone could fit both a ballad and a dance song.

Can I use my own lyrics with AI song prompts?

Yes. Write the lyrics yourself and keep the prompt for the sound: genre, mood, instruments and voice. In our study, 30% of songs were made from lyrics the user wrote, which also avoids reusing lines from existing songs.

What words should I leave out of a song prompt?

Leave out artist names, band names and lines from existing songs, since most models refuse or rewrite them. Skip vague words like good or catchy too, because they give the model nothing to work with. Describe what a listener would hear instead.

How many instruments should a song prompt name?

Name two or three lead instruments and how they are played, like fingerpicked acoustic guitar, warm Rhodes piano and brushed drums. A long list blurs the mix, while two or three clear parts give the track a distinct character.

Why does the same prompt give a different song each time?

AI song models generate a fresh take on every run, so the same prompt produces different melodies. A detailed four-ingredient prompt keeps the sound consistent across takes even when the melody changes. Generate a few versions and keep the best one.

Explore more

AI Song Generator

Lyrics to Song

We Analyzed 6,798 AI Song Generations: 1 in 10 Prompts Copies a Real Artist’s Work

AI Mastering

AI Stem Splitter

AI Vocal Remover

Write your four ingredients and hear the result.

Why AI song prompts that name a singer get refused or sound flat, and the 4-ingredient recipe that works instead: genre, mood, instruments and voice.