TL;DR
Making a song with AI is five moves: describe the song (or write its lyrics), generate two takes, listen and keep one, refine it with the editing tools — extend it, cover it in another style, replace a weak section, pull the vocal off — and use it under a licence you actually hold. On Suno AI Music a new account starts with 16 free credits, a generation costs 10 and returns two complete songs, and everything you make lands in your library where the other tools can pick it up. This guide walks the whole path once, then points to the deeper guides for each step.
In this guide
- What AI music generation does
- Step 1: Decide what the song is
- Step 2: Describe it, or write it
- Step 3: Generate two takes
- Step 4: Refine what you kept
- Step 5: Use it with the right licence
- What it costs
- FAQ
What AI music generation does
A music model is trained on a very large amount of audio paired with descriptions. Given a new description — genre, mood, instruments, a voice — it composes and performs a complete recording: melody, arrangement, production and singing, in one pass. It is not stitching loops, and it is not pulling an existing song from a database; each generation is new audio.
Three things follow that shape everything below:
- The description is the instrument. You do not play notes; you describe a record. Learning to describe precisely is the whole skill.
- Every run is a draw. The same prompt gives different songs each time, which is why the generator returns two takes and why iterating is cheaper than perfecting a prompt in advance.
- The output is a finished recording, not a project file. You get a mixed stereo track. To change it you generate again, or use the tools built for editing generated audio — extend, cover, replace, separate — rather than opening a DAW.
Step 1: Decide what the song is
Ten seconds of thinking here saves ten credits later. Answer three questions before touching the generator:
- What is it for? A TikTok bed needs 30 seconds and a hook; a podcast intro needs a clear ending; a song for its own sake needs a chorus that earns repeats.
- What does it sound like? One genre, one mood. If you cannot name them, name a record it should sit next to on a playlist — then describe that record's sound rather than its artist.
- Does it have words? Instrumental is a switch on the generator and is the right answer for most background uses. If it has words, decide whether you are writing them or letting the model write them.
Step 2: Describe it, or write it
Open the create page. It has two modes.
Simple mode is one description field, up to 500 characters. Write the sound and the subject together in a sentence or two — the page's own example is "A dreamy lo-fi beat about rainy nights in Tokyo, with soft piano and vinyl crackle." The model writes the lyrics and picks the arrangement. Use it for a first draft, for background music, or whenever the words are not the point.
Custom mode separates the job: a Style field for the sound (up to 1,000 characters), a Lyrics field for the words (up to 5,000), a Title, and two extras — Negative Tags for things you do not want, and Vocal Gender. Use it when you have written lyrics, or when Simple mode keeps getting the sound almost right and you need to steer one thing at a time.
Two rules that hold in both modes:
- Style goes in the style box; story goes in the lyrics. "A song about my grandmother" is not a sound. Genre, mood, instruments and vocal are.
- Use the four-part formula for style: genre, mood, instrumentation, vocal —
acoustic folk, tender, fingerpicked guitar, close-mic female vocal. The Suno prompts guide has thirty ready-made examples by genre.
If you want your own words but do not want to write them from scratch, the AI lyrics generator turns a two-sentence brief into a full sheet with [Verse] / [Chorus] tags for 1 credit, and a button sends it straight into Custom mode.
Toggle Instrumental for no vocals. Pick a model: Suno v4.5 is the free one, v5 and v5.5 are PRO and come with the paid plans.
Step 3: Generate two takes
Click Generate Music. One run costs 10 credits and produces two songs, each a complete track from the same prompt. They land in the songs panel beside the form and in My Music, usually within a couple of minutes.
Listen to both all the way through before judging. Then decide which of three things happened:
- One take is right. Keep it. Move to step 4 if it needs work, or step 5 if it does not.
- Both are close but not right. Change one slot of the style prompt — the genre, the mood, the vocal — and run again. Two runs that each change one variable teach more than five that change everything.
- Both are wrong. The prompt has a contradiction or the story is in the style box. Rewrite the prompt with the four-part formula before spending more credits.
Step 4: Refine what you kept
This is the part beginners skip, and it is where a generation becomes a song. Every tool below works on any song in your library and costs the same 10 credits as a generation.
- Too short, or the ending is weak? Extend it. Drag the Continue At slider to the last bar you like and the model writes onward from there — the original stays byte-for-byte, and you get two full-length versions. Guide: how to extend a song.
- Right song, wrong record? Cover it in another style. Melody and lyrics stay; arrangement, production and voice are re-performed. Guide: how to make an AI cover.
- One bad section in a good song? Replace it. Mark the stretch and the model rewrites only that, keeping everything on either side.
- Need the instrumental, or just the voice? Separate the vocals. One run returns a clean instrumental and an isolated acapella. Guide: how to remove vocals online.
Because every result is saved as its own song, the tools chain: extend a cover, cover an extension, separate the result. Keep the earlier versions; they are your undo.
Step 5: Use it with the right licence
Everything you generate on the site is copyright-free — no sample to clear, no artist to credit. What changes with the plan is what you may do with it:
- The free tier is for making and listening: 16 credits on signup, Suno v4.5, your songs in your library.
- The paid plans add a monthly credit allowance, the PRO models, high-quality downloads, and the commercial licence — with a certificate you can download — that covers use in videos, podcasts, ads, streaming platforms and client work. Covers, extensions and stems inherit the licence of the song they came from.
If the song is going anywhere public, read the licence once before you upload it; the rules platforms apply to AI music are about disclosure and ownership, and a certificate answers both.
What it costs
| Action | Credits | You get |
|---|---|---|
| Generate a song | 10 | Two complete takes |
| Extend / cover / replace | 10 | Two takes of the edited song |
| Separate vocals | 10 | Instrumental + acapella |
| Write lyrics | 1 | A full tagged lyric sheet |
A new account's 16 credits are one generation plus six lyric drafts. Every action above draws from the same balance, and the plans are priced by monthly credits — pick the one that matches how many songs a month you actually finish, not how many you might start.
FAQ
Do I need to know music theory?
No. The model handles key, harmony and arrangement. What you need is the vocabulary to describe a sound — genre names, moods, instrument names — and the patience to iterate.
How long does a generation take?
Usually a couple of minutes for two takes; jobs run in the background, so you can leave the page and come back to them in My Music.
Can I upload my own audio?
The tools here work on songs in your library — generated, extended, covered or separated on the site. There is no upload of outside audio, which is also what keeps every result cleanly licensed.
Can I use AI music commercially?
On the paid plans, yes, with a downloadable certificate. Free-tier songs are for personal use. See the plans for what each tier covers.
What is the difference between Simple and Custom mode?
Simple is one description field and the model writes the lyrics; Custom splits style, lyrics and title into their own fields and adds negative tags and vocal gender. Draft in Simple, control in Custom.
Which model should I pick?
Start on v4.5 — it is free, and it is what a free account uses regardless. Switch to v5 or v5.5 when you want the newer sound and have a paid plan; every model costs the same 10 credits.
Where do my songs go?
To My Music, where you can play, favourite, share, download (on paid plans) and hand them to any of the editing tools.
Final take
Describe a record, not a feeling; generate two, keep one; fix the part you don't like with the tool built for that part; and read the licence before the song goes public. That is the whole method, and the first run is free.
Start here: make a song.