Write lyrics the model actually follows.
Square-bracket tags tell the model what each block of lyrics is for, the Style box sets the sound, and only singable words go in between. This guide is the distilled version of what the best Suno and AI-music tutorials agree on.
Tags are read. Lyrics are sung.
Type structure tags like [Verse] and [Chorus] on their own lines in the Lyrics box — the model reads them as instructions instead of singing them. Every unbracketed line is meant to be sung as written. The sound itself — genre, tempo, vocal, mood, instruments — belongs in the Style box, not in the lyrics. Generate, then fix single sections instead of rewriting the whole song.
LYRICS — what gets sung
Every unbracketed line is performed as written: your words, your rhymes, your imagery.
TAGS — what gets read
Bracket lines set the job of each block — structure, delivery, breaks. The model treats them as strong hints, not commands.
Write a song in six moves
Plan the structure, tag it, split the fields, then iterate one section at a time.
- 1
Plan the shape before the words
Decide the sections first. A typical three-minute song runs Intro → Verse 1 → Chorus → Verse 2 → Chorus → Bridge → Final Chorus → Outro. Architecture first, words second. Draft the plan wherever it is easiest — paper, notes, or a chat AI — then paste the tagged result.
- 2
Tag every section in square brackets
In the Lyrics box, put [Intro], [Verse 1], [Pre-Chorus], [Chorus], [Bridge], [Outro] on their own lines, each right above the lines it governs — [Build], [Drop], and [Breakdown] fit dance styles. One tag per line, and write tags in English even when the lyrics are in another language.
- 3
Write only singable words between the tags
Every unbracketed line gets sung, so keep chorus lines short and repetitive, match syllable counts between parallel verses, and leave planning notes out — unbracketed instructions get performed.
- 4
Describe the sound in the Style box
In Custom mode, describe the sound in the Style box: genre first, then tempo, vocal type, mood, and instruments — the most important words first, comma-separated, no square brackets here. Example: dream pop, mid-tempo, airy female vocal, warm synth pads, bittersweet.
- 5
Add local cues only where the plan changes
A short instruction inside a tag — [Bridge - drums fall away, close vocal] — changes one section without touching the rest. Parentheses cue the delivery of a single line: (whispered), (spoken), (harmony).
- 6
Generate, cherry-pick, fix sections
Each run returns two takes. Keep the better one, then repair single sections with Replace Section or continue the song with Extend — reusing the exact style wording — instead of regenerating everything from scratch.
How to write tags the model follows
Tags are strong hints, not commands. Simple, consistent, front-loaded tags get followed most.
Do
- One tag per line, on its own line, right before the section it governs.
- Use the same name every time — if the chorus is [Chorus], never switch to [Hook] halfway through.
- Write tags in English even when the lyrics are in another language.
- Put the most important instruction first — early words carry the most weight.
- Number the sections and repeats: [Verse 2], [Final Chorus] — and cue the ending with [Outro], [Fade Out], or [End] so the song lands instead of rambling.
Avoid
- Stacking contradictory tags on one section — [Chorus][Drop][Whispered] fight each other.
- Repeating the whole style description beside every section. Global sound belongs in the Style box only.
- Writing plain-prose directions between the lines — “now make it sadder” gets sung.
- Expecting 100% obedience. If a tag is ignored, simplify the wording and generate again.
- Regenerating the whole song because one section failed.
How to write lines that sing well
The model performs what you type. Lines written for the ear land; lines written for the page get rushed.
Do
- Keep chorus lines short, simple, and repetitive — repetition is what makes a chorus feel big.
- Match syllable counts between parallel lines so the verses share a melody.
- Write the chorus out in full under every [Chorus] tag, with the same wording, when you want an identical return.
- Use concrete images and plain words — “cold kitchen light” sings better than “ephemeral solitude”.
- Leave room to breathe: a short line or an [Instrumental Break] gives the arrangement space.
Avoid
- Dense literary lines — long words get rushed or mumbled to fit the bar.
- Letting the chorus wording drift between repeats, so every chorus gets a new melody.
- Filling every second with words. Songs need air.
- A wall of untagged text — the model guesses the structure and usually guesses short.
- Forcing a rhyme at the cost of the natural phrase. Near-rhymes sing better than broken grammar.
How to write a Style prompt that sounds right
The Style box is a short, comma-separated description of the sound. Order carries weight — the words you put first shape the song the most.
Do
- Lead with genre and era — "late-90s alt rock" tells the model more than ten adjectives.
- Name the vocal once and concretely: "airy female vocal" or "gravel male baritone", not "great singer".
- Add production texture — "dry drums", "warm analog synths" steer the mix, not just the notes.
- Stay around 5 to 8 elements — every extra word dilutes the ones before it.
- Reuse the exact same style wording across runs while you iterate — it is your control variable.
Avoid
- A wall of 20 keywords — everything competes and the model picks at random.
- Contradictions — "fast tempo, slow ballad" resolves to a coin flip.
- Vague praise words — "beautiful, amazing, emotional" say nothing about sound.
- Structure directions in the Style box — "[Chorus]" and "verse 2 should be sadder" belong in the Lyrics box.
- Rewriting the style between attempts — you can no longer tell what actually changed.
The tag cheat sheet
Every tag worth knowing, grouped by job. All of them go on their own line, in English, right above the section they govern.
Structure — the skeleton
[Intro]Short opening. Sets the mood before the first verse — often instrumental or a slow reveal.
[Verse]The story sections. Number them — [Verse 1], [Verse 2] — so the model treats them as a set.
[Pre-Chorus]A short lift between verse and chorus. Builds tension so the chorus lands harder.
[Chorus]The hook. Biggest, simplest, most repeated section — keep the wording identical on every return.
[Bridge]A deliberate departure before the final chorus. New musical ground, a new angle on the lyric.
[Outro]The closing section. Cue it so the song ends on purpose instead of stopping mid-thought.
[Fade Out] / [End]Ending instructions. [Fade Out] trails away; [End] stops clean.
Arrangement & energy
[Build]Rising tension: layers and energy stack toward a peak. Common in dance and pop.
[Drop]The release after a build. Everything hits at once — beat, bass, hook.
[Breakdown]Strips the arrangement to a few elements. Quiet and spacious, usually before the final chorus.
[Instrumental Break]A stretch with no vocals. The band or the synths take the lead.
[Guitar Solo] / [Piano Solo]A featured solo. Swap in the instrument your style calls for.
Vocals & delivery
[Whispered]Sung soft and close. Works inline too: (whispered) before a single line.
[Spoken]Talked, not sung. Good for intros, interludes, and spoken-word moments.
[Harmonies]Layered backing vocals around the lead line.
[Female Vocals] / [Male Vocals]Sets the vocal type for one section — or set it once, globally, in the Style box.
(whispered), (spoken), (harmony)Parenthetical cues affect one line only — and weigh less than section tags.
Need a one-off change? Put a short instruction inside the tag itself — [Bridge - drums fall away, close vocal]. One section, one instruction, nothing global.
Easy to get wrong
Untagged lyrics leave the structure to chance: the model picks its own section lengths and may stop after a single verse.
Anything not in brackets is sung. Keep production notes and plans like “verse 2 should be sadder” out of the Lyrics box.
Square brackets in the Style box are the wrong syntax — the Style field is plain comma-separated words.
Parentheses are for delivery cues on one line — (whispered), (spoken), (harmony) — not for background text.
Position carries weight: words at the top of a field matter more than words at the bottom, in both boxes.
If the model ignores a tag, don't stack more tags — simplify the wording and generate again.
Iterate one change per run. New style plus new lyrics plus new structure tells you nothing about what actually worked.
An instrumental section is a tag: [Instrumental Break] or [Guitar Solo] on its own line, or an empty section under a tag.
Tagged lyrics vs plain text
Same words, same style prompt — the bracket tags are the only difference.
| If you want to… | Tagged lyrics | Plain text |
|---|---|---|
| Set the song length | Add or repeat tagged sections | Length is less predictable |
| Bring the chorus back identically | Repeat the same tagged chorus text | Repeats tend to drift |
| Fix one bad section later | A tagged [Verse 2] is easy to target with Replace Section | Hunt for the span by ear |
| Control where vocals enter | [Intro] and [Instrumental Break] hold them back | Vocals start right away |
| Be surprised by the arrangement | Tags constrain the outcome | Happy accidents happen |
Set the song length
- Tagged lyrics:Add or repeat tagged sections
- Plain text:Length is less predictable
Bring the chorus back identically
- Tagged lyrics:Repeat the same tagged chorus text
- Plain text:Repeats tend to drift
Fix one bad section later
- Tagged lyrics:A tagged [Verse 2] is easy to target with Replace Section
- Plain text:Hunt for the span by ear
Control where vocals enter
- Tagged lyrics:[Intro] and [Instrumental Break] hold them back
- Plain text:Vocals start right away
Be surprised by the arrangement
- Tagged lyrics:Tags constrain the outcome
- Plain text:Happy accidents happen
Lyrics FAQ
Lines in square brackets — [Verse], [Chorus], [Bridge] — written on their own lines in the Lyrics box. They tell the model what each block of lyrics is for. The model reads them as instructions and does not sing them — though malformed or unusual tags can occasionally leak into the vocals, one more reason to keep tags simple.
Yes. Write tags in English even for Chinese, Japanese, or any other lyrics — the models were trained on English tags. The lyrics themselves can be in any language.
Because every unbracketed line is treated as lyrics. Put structure in bracket tags and keep planning notes out of the Lyrics box entirely.
In the Style box, as plain comma-separated words — no brackets. Genre and tempo first, then vocal type, mood, and instruments, with the most important words first.
Yes, with a caveat: parenthetical cues are weaker than section tags and sometimes get ignored. Parentheses right before a line cue the delivery — (whispered), (spoken), (shouted), (harmony). For a whole-section change, put a short instruction inside the section tag instead: [Bridge - stripped back, close vocal]. Section tags carry more weight.
A typical full song runs two verses, two or three choruses, and maybe a bridge — roughly 30 to 60 sung lines for three minutes: fewer for sparse pop, more for rap. More tagged sections nudge the song longer; for reliable extra length, extend a finished take instead. The model stretches or trims delivery to fit, and may repeat or paraphrase lines at the extremes.
If you write the chorus once and mark the later repeats [Chorus], the model may vary the melody. The most reliable fix is pasting the same chorus text under every [Chorus] tag — identical wording makes an identical return far more likely, though still not guaranteed.
Tag it. [Instrumental Break], [Guitar Solo], or [Piano Solo] on its own line gives the arrangement a stretch without vocals. An empty section under a tag also leans instrumental.
Tags are strong hints, not commands. Simplify the wording, move the key instruction to the front, drop the stacked cues, and generate again — a fresh take usually follows a cleaner prompt.
Extend continues after a cut: give the new part one clear tag like [Final Chorus] and reuse the original style wording. Replace Section rewrites one span, and the tags already in your lyrics are the anchors it edits between.
Related guides
Plan it. Tag it. Generate.
Open Create, sketch your sections in square brackets, and put the sound in the Style box — the structure will hold.