How to make a song with AI, from a one-line idea to a finished track
Making a song with AI is not one action. It is a brief, a lyric, a musical description, and then a few rounds of judgement about which of several takes is actually good. People who get useful results treat it as four steps; people who type a sentence and take whatever comes back get the results that make them think the tools do not work.
The four steps below are the whole process.
Step 1: write the brief before you write anything else
Answer four questions in one line each. This takes two minutes and changes everything downstream.
- Who is singing, and to whom? A song with no addressee drifts into generalities.
- What is the single feeling? One. Songs that carry two compete with themselves.
- What is the setting? A concrete place gives the lyric its images.
- Where does it end up? Resolved, unresolved, or somewhere worse than it started.
A filled-in brief looks like: someone leaving a city they loved, to the city itself, wistful but not bitter, late summer, ends accepting it.
A brief like this is more useful than any amount of prompt engineering, because everything after this is downstream of it.
Step 2: the lyric
You can generate lyrics or write them. Both work; the difference is how much of the song is yours.
If you generate, give the brief rather than a topic. "Write a song about leaving" produces greeting-card output. The brief above produces something with images in it.
Structure matters more than people expect. Ask explicitly for the shape:
Verse, chorus, verse, chorus, bridge, chorus. Keep the chorus identical each time. Concrete images, no abstractions like "forever" or "destiny".
The instruction against abstractions is the single most effective line in lyric prompting. Generated lyrics default to abstraction, and abstraction is what makes them sound generated.
Our songwriting guide goes deeper on structure and revision, and the free lyric generator is a starting point if you would rather begin from something than from nothing.
Step 3: the musical description
The musical description is where most of the sound is decided. Name four things:
- Genre and era. Late-nineties alternative rock, not rock. Era carries production style.
- Instrumentation. Two or three specific instruments. Fingerpicked acoustic guitar, brushed drums, upright bass.
- Tempo and feel. Slow, unhurried, or a rough BPM if you know it.
- Vocal. Range, texture, delivery. Warm low female vocal, close-mic'd, restrained.
A complete description:
Late-nineties alternative rock, fingerpicked electric guitar and brushed drums, mid-slow tempo, warm low female vocal, restrained delivery, building only in the final chorus
Vocals are available across 35 languages, so the language of the lyric is a choice rather than a constraint. Our genre guide covers which descriptions produce the most consistent results.
Step 4: generate several, then judge
Judging several takes is the step people skip. A track costs 25 credits. Generate three or four from the same brief and lyric before deciding anything.
Judge them in this order:
- Does the chorus land? If the chorus is not the best part, nothing else matters.
- Is the vocal intelligible? Play it once without reading the lyric.
- Does the arrangement change across the song? A track that is identical from start to finish is a loop, not a song.
- Would you listen to it twice? The only question that actually predicts whether anyone else will.
Take the best one. If none pass, change one thing in the description — usually the vocal or the tempo — and generate again. Changing everything teaches you nothing.
When the track is finished
A generated track is a recording, not a project. Finishing usually means:
- Trimming the intro or outro to the length you need.
- Pairing it with visuals — the lyric video guide covers the most direct route, and the music video guide covers the ambitious one.
- Making cover art at the right shape, with room for the title — our album cover guide covers designing the empty space first.
The rights question, answered plainly
You can use tracks you generate. What nobody can honestly give you is an unconditional legal guarantee, because the copyright status of AI-generated music is still developing and varies by jurisdiction. In the United States, the Copyright Office published its report on the copyrightability of generative-AI outputs in January 2025 (copyright.gov, "Copyright and Artificial Intelligence", checked November 2026), while UK law has its own rule that treats the author of a computer-generated work as the person who made the arrangements for its creation (legislation.gov.uk, Copyright, Designs and Patents Act 1988, section 9, checked November 2026). Treat any tool promising otherwise with suspicion.
Our licensing explainer sets out the real position, and can you use AI music on YouTube covers how platform matching systems behave in practice, which is a separate question from the legal one.
If music is the part you are working on, these go further: How to make a beat with AI, How to make a jingle with AI, Song structure explained, AI music for podcasts, and Music for Twitch streams.
Frequently Asked Questions
Can AI write both the lyrics and the music?
Yes — you can generate lyrics from a brief and then generate a full performance from those lyrics plus a musical description. The results are better when you write or heavily edit the lyric yourself, because the lyric is where a song is most obviously human or not.
How long does it take to make a song with AI?
The generation takes minutes. Getting a track you would actually release takes an hour or two, mostly spent writing a better brief and rejecting takes. The generation is not the slow part and never was.
What makes AI lyrics sound generated?
Abstraction. Words like forever, destiny, shadows, and journey, used without a concrete image attached. Instruct explicitly against them and ask for specific objects, places and actions instead — the difference is immediate.
Can I make a song in a language other than English?
Yes. Vocals are supported across 35 languages. Write or generate the lyric in that language rather than translating an English one, since translated lyrics rarely scan or rhyme after the fact.
How much does an AI song cost?
25 credits per full track. Budget for three or four attempts rather than one, because choosing between takes is most of what makes the final track good. A free account starts with 150 one-time credits, which covers a first song with room to reject a few versions.
Write the brief and generate four
Spend two minutes on the four brief questions, generate four takes, and keep one. That is the whole method, and the first time through it takes an evening. Open the Music Studio — a free account includes 150 one-time credits with no card required.
Not sure which generator to start with? The best AI music generators in 2026 sorts seven of them by what the track is for.
Written with AI assistance. Ryan Balazadeh, founder of PrismPoster, is responsible for this article and its corrections.


Produce original AI music with custom vocals for your videos and streams
Create complete custom tracks with vocals in 35 languages, extract stems, generate cover art, and export finished songs.