A birthday song
The first name in the chorus, two memories in the verse, a style that sounds like the person. The gift that costs under a dollar and gets played again ten years later.
Write a song with AI
The same lyrics sung twice by ElevenLabs Music: a bright pop track and an acoustic folk take, one minute each. The full workflow is visible without an account, with both songs.
Interactive demo
Real workflow, real results. Pan and zoom freely.
Use this workflowReal workflow: two Text nodes, each holding the style and the same lyrics, wired into two Music nodes on ElevenLabs Music. Listen to both versions.
At a glance
- What the workflow produces
- 2 audio tracks, 120 s in total
- Canvas structure
- 5 nodes wired by 2 connections
- Models wired in
- ElevenLabs Music
- Cost of one full run
- about 168 credits, roughly $2.02
A song is words that someone sings. Music generators have known how to make a melody for a while; what changed is that they now sing your lyrics, in your language, without rewriting them. This workflow shows it on an eight-line text: a verse and a chorus about a road at dawn, sent to ElevenLabs Music twice with two style prompts.
Each prompt has two parts, and the order matters. First the style, in one sentence: genre, voice (female or male), instruments, tempo, mood. Then the lyrics, introduced by "sing exactly these lyrics". Without that sentence, the model treats your lines as inspiration and invents its own; with it, it sings them in order. The demo lets you hear it: the pop and the folk sing the same chorus, word for word.
Sixty seconds cost the same as thirty, because the provider bills by the started minute: always ask for at least a minute. The two versions of the demo come to two dollars, eighty-four credits each. That is the price of a whole song, with the voice, the arrangement and the mix.
How it works
- 1
Write the lyrics
A verse and a chorus are enough for one minute. Label them, "Verse:", "Chorus:", the model reads the structure. Short lines sing better than long sentences.
- 2
Describe the style in one sentence
Genre, voice, instruments, tempo, mood: "upbeat pop, female vocal, bright synths, 120 bpm". The model picks whatever you leave out.
- 3
Paste "sing exactly these lyrics"
Between the style and the lyrics. It is the sentence that turns your text into sung lyrics rather than a theme. Duplicate the Text node for a second style: the lyrics stay, the style changes.
- 4
Listen to both, keep one
"Generate all" runs both songs. Listen to the chorus first: it is what people remember. Keep the version you like, or change one word of the style and rerun that node alone.
Variations and use cases
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
A jingle for a brand or a podcast
Fifteen seconds of chorus are enough, but ask for the minute: it costs the same, and you will trim it in the edit. The brand name in the lyrics, the audience's taste in the style.
A demo of a track
Your lyrics, three different styles, one minute each: a songwriter hears the text sung before choosing an arrangement. A demo convinces a musician better than a text file.
A song for a video
The lyrics tell the video, the style follows its pace. The Music node then wires into the Edit node, which lays the song under the shots and cuts it at the end.
A nursery rhyme for a child
Very short lines, simple rhymes, a chorus that repeats the first name, a style like "gentle lullaby, female voice, guitar and xylophone, slow". The model knows how to slow down.
The song, then the voice isolated
Once the song is generated, the separate the vocals from a song workflow pulls out the sung track alone, to remix it or lay it on another arrangement.
Common mistakes
Forgetting "sing exactly these lyrics"
Without that sentence your lines become a theme and the model writes its own words. It is the most common mistake, and the easiest to fix.
Asking for thirty seconds to pay less
The provider bills by the started minute: thirty seconds cost the same as sixty. Ask for the minute, cut afterwards if needed.
A style without a voice
"Upbeat pop" does not say who sings. State "female voice" or "male voice", and the language: the model decides at random whatever it is not told.
Lyrics too long for the duration
Four verses in one minute give a rap delivery on a ballad. One verse and one chorus per minute, the chorus sung twice, is the right ratio.
Recommended models
ElevenLabs Music
DemucsFrequently asked questions
Does the model really sing my lyrics?
Yes, when the prompt says "sing exactly these lyrics". Both songs in the demo repeat the chorus word for word. Without that sentence, it draws inspiration from them freely.
Can I get the version without the voice?
Yes. The Music node has an "instrumental" switch: on, the model produces the arrangement alone from the same prompt. That is the version to sing over or to put under a video.
Which languages does the model sing in?
The language of the lyrics. Write them in English, Spanish or Japanese and state the language in the style: "vocal sung in English". The demo is in French.
How much does a song cost?
Eighty-four credits per minute, about a dollar, whether you ask for thirty or sixty seconds. Two minutes cost twice that. The price shows on the node before you click.
What is the maximum length?
Three minutes per node, in steps: thirty seconds, one, two or three minutes. For a longer track, chain two nodes and assemble them in the Edit node.
Can I use the song in a commercial video?
Generated music is yours to use under ElevenLabs' terms on a paid account, which is what runs here. Check the lyrics: they must be yours, not those of an existing song.
Can it imitate an artist?
No, and the provider refuses the request. Describe a style, an era, instruments, a vocal timbre: "deep raspy voice, sixties blues". Not a name.
Can I do it from a phone?
Yes. The canvas runs on mobile, the lyrics are typed in the Text node and the song plays in the Music node. The create on mobile guide explains the gestures.
Write a song with AI
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.