Every Suno metatag that actually works (lyrics and structure tags)

Paste this into the Lyrics box exactly as written, (soft whisper, building to a scream), and hit generate. Suno will not whisper. He will sing the words "soft whisper building to a scream" in a clear, confident voice, right in the middle of your verse, and you will have spent a generation figuring out why. Brackets are the fix. [whispered] on its own line tells the singer how to sing without ever becoming a lyric. That single distinction, what gets sung versus what gets obeyed, is the whole logic of metatags, and most wasted credits come from getting it backwards.

The rule the whole system rests on

Anything you type into the Lyrics field goes to the microphone unless it sits inside square brackets. [Chorus], [whispered], [Guitar solo] are instructions, and they stay silent. I never meant to stay this long is a lyric, and it gets sung. There is a second, quieter bracket in play. Round parentheses ( ) are usually read as lyrics too, specifically as backing vocals or ad-libs, so (ooh, ooh) under a line will often get sung by a second voice. So the working split is simple: square brackets talk to the model, parentheses add a harmony part, plain text goes to the lead vocal.

Here is the mistake almost everyone makes on their first Custom song. A writer types the sections out like a document:

Verse 1:
I walked the city half asleep
Chorus:
And you were every light I passed

Suno dutifully sings "verse one" and "chorus" as if they were lines, because the words sit outside brackets, so they are lyrics. Wrap them, [Verse] and [Chorus], each on its own line, and the narration vanishes while the arrangement snaps into shape. That one edit is the difference between a demo you delete and a demo you keep.

Structure tags: the skeleton of the song

These mark where sections begin and end. They cost nothing and they carry more weight than any other group, because a song with labelled sections re-rolls far less often than a shapeless one. On the free tier every generation spends part of a daily credit allowance, so a prompt that drifts and forces a re-roll is real money and real minutes. Structure tags are the cheapest insurance you can buy against that.

• [Intro]: an opening, often instrumental, before the first vocal enters.
• [Verse]: the narrative body. Number them, [Verse 1], [Verse 2], when you want distinct melodies rather than a repeat.
• [Pre-Chorus]: the two or four lines that build tension and hand off into the hook.
• [Chorus]: the hook, the repeating part. Put your strongest line here, because this is what the listener leaves with.
• [Bridge]: the contrast section, usually about two thirds of the way in, a change of chord or mood that stops the song feeling like a loop.
• [Outro]: how it ends. Pair it with [fade out] when you do not want a hard stop.

Arrangement and instrumental tags

These shape what the band does rather than what the singer does.

• [Instrumental]: no vocals in this stretch, just the arrangement.
• [Guitar solo]: a lead guitar break. You can swap in [Piano solo] or [Saxophone solo], though the more exotic the instrument, the less reliably it lands.
• [Build]: rising tension, filters opening up, common in electronic and cinematic styles.
• [Drop] or [Beat drop]: the release after a build, where the bass and the beat hit together.

Be honest with yourself about genre fit. [Build] into [Drop] works reliably in EDM-flavoured styles and gets quietly ignored in a folk ballad, because there is nothing there to drop. The tag is not broken. It just has nothing to grab in that arrangement.

Vocal and performance tags

These steer how a line is delivered.

• [whispered]: breathy, quiet, intimate delivery.
• [belted]: full chest, loud, the big note you build the chorus around.
• [harmonies]: stacked backing vocals under the lead.
• [spoken word]: talked rather than sung, good for an intro monologue or a bridge.
• [falsetto]: light head voice, high and airy.

Treat these as hints, not switches. [belted] on a sleepy lo-fi track gets softened, because the Style field wins ties. If the whole style says "quiet late-night bedroom pop" and one tag says "belt it", the style usually holds the line. When a vocal instruction really matters, reinforce it in both places: the tag in the Lyrics and the matching descriptor, powerful belted female vocal, in the Style field.

Dynamics and effects

• [fade out]: a gradual drop to silence at the end, instead of a clean cut.
• [fade in]: the reverse, easing up from nothing at the start.
• Less reliable, worth a try but do not build a song around them: [key change], [tempo up], [double time]. Newer model versions honour these more often than older ones did, but none of them promise to.

The uncomfortable truth: there is no official list

Suno has never published a complete, guaranteed catalogue of tags. What circulates is a community map built from trial and error, and it is genuinely useful, but it is not a spec. Some tags work every single time, [Chorus], [Verse], [Instrumental]. Some work in one genre and evaporate in another. Some get silently ignored. And a few get sung by accident when the model does not recognise the word inside the brackets as an instruction it knows. Newer versions, v4.5 and v5, follow structure tags more faithfully than v3.5 did, and v4.5 widened the range of styles the tags behave inside, but treat any exact per-version claim as something to verify against the current release notes rather than a two-year-old forum thread.

What that means in practice: never assume a tag worked because you typed it. Listen for it. If you cannot hear the difference, it did not fire.

How to test whether a tag actually works

One variable at a time. This is the only method that gives you a clean answer, and it takes two generations.

1. Write a short Lyrics block, two sections, nothing fancy.
2. Generate once with the tag in place. Note exactly what happens at that spot.
3. Generate again with everything identical, the same Style, the same lyrics, only the tag removed.
4. Compare the two takes. If they differ in the way the tag promised, the tag works in that style. If they sound the same, the tag is being ignored, and you should move the instruction into the Style field instead.

The discipline is in step 3: keep the Style field, the seed if you are reusing one, and every other line frozen. Change only the tag. If you change the genre and the tag in the same run, you learn nothing, because you cannot tell which edit moved the sound.

Quick reference table

A full example: Russian lyrics, English tags

The structure and delivery tags stay in English because Suno reads them most reliably in English. Only the sung lines follow the song's language. Here the words are Russian and every bracket is English.

[Intro]
[spoken word]
Город не спит, и я вместе с ним

[Verse]
Фонари роняют свет на мокрый асфальт
Я считаю шаги до твоих дверей

[Pre-Chorus]
[build]
И воздух становится тяжелее

[Chorus]
[belted]
Останься со мной до утра
Пока не погаснет последний вагон

[Bridge]
[whispered]
Я не умею иначе

[Outro]
[fade out]
Останься

Notice the sung lines are short and use simple, open vowels. Suno can mispronounce Russian when a line is long or consonant-heavy, so short phrases and clean vowels cut the risk. The matching Style field for this would be something like melancholic Russian indie pop, rainy late-night mood, live drums, warm synth pads, emotive female vocal, mid-tempo. If staring at an empty Style box is the part that actually stalls you, a catalog of ready-made Suno prompts like sunomarket gives you a tested phrase to paste and adjust instead of guessing at descriptors.

FAQ

Do capital letters or the exact spelling matter inside a tag? Case is forgiving. [chorus] and [Chorus] behave the same. Spelling matters more, because the model is matching a known concept. [Chorus] is understood everywhere, an invented synonym may not be. When a tag is being ignored, try the plainest, most common wording before you conclude the feature does not exist.

Will Suno sing my section labels if I forget the brackets? Yes, and this is the single most common beginner bug. Chorus: typed as plain text is a lyric, so the model sings the word "chorus". Anything outside square brackets goes to the microphone. If you hear the singer narrating your structure, missing brackets are almost always the cause.

Can I invent my own tags like [explosion] or [angry]? You can type anything, but there is no guarantee the model does anything with it. Emotional and sound-effect tags are hit or miss. [angry] sometimes shifts the delivery, sometimes does nothing, and occasionally gets sung. Test it with the one-variable method before you rely on it, and if it fails, push the intent into the Style field where descriptors like aggressive, distorted, shouted vocal are read more consistently.

Why did my [Guitar solo] come out as vocals? Usually because a lyric line sat directly under it. If there is text below the tag and outside brackets, the model treats that text as words to sing over the section. For a purely instrumental break, put [Guitar solo] or [Instrumental] on its own and leave no lyric lines beneath it until the next section tag.

How many tags is too many? Enough to mark structure and a few key delivery moments, no more. A tag on every line fights the model and often gets averaged away. A clean skeleton, [Intro] [Verse] [Pre-Chorus] [Chorus] [Verse] [Bridge] [Chorus] [Outro], plus two or three vocal hints where they genuinely matter, outperforms a lyric sheet drowning in brackets almost every time.