Suno v6 syntax, condensed

Current as of the September 2026 v6 launch. Three models: v6 (control), v6-wild (discovery), v6-mini (fast, free tier). All three generate up to 8 minutes.

Two fields, two jobs

• Style box: how it sounds. Nothing here is sung. Comma or period separated prose, no brackets. Order by influence: genre → mood → lead instrument → vocal → production. 3–5 tags for simple songs, up to ~15 for control, beyond ~20 the model gets confused.
• Lyrics box: what is sung, plus [bracket tags] that shape the arrangement. Tags are case-insensitive and probabilistic cues, not commands.
• Exclude styles (Advanced options): first-class negative field. More reliable than inline no drums; combine both for stubborn elements.

Section tags that are validated

[Intro] [Verse] [Verse 2] [Pre-Chorus] [Chorus] [Post-Chorus] [Hook]
[Bridge] [Break] [Breakdown] [Build] [Drop] [Interlude]
[Instrumental] [Solo] [Outro] [End] [Fade Out]

Pipe stacking (the v5.5/v6 pattern)

[Section | Mood | Vocal | Instruments | Movement | Space | Production]
[Chorus | explosive | full voice | wide synths | stacked harmonies | vocal forward]

Max ~7 elements, most important first, never contradictory pairs in one tag. Stacked tags override the Style box for that section.

Typography the vocal model reads

LINE IN CAPS!        louder, shouted (use once, for contrast)
line in lowercase     delicate
(echo these words)    backing vocal / call-and-response
[yeah!] inside a line adlib, not a section
lo-o-o-ove            stretched note
i-hit-the-street      hyphen run = faster flow
word, word.  word...  breath · hard reset · drag
*rain falling*        non-sung atmosphere
reed / lyve / bahss   phonetic respelling for homographs
[End]                 terminal tag, ~85–90 % effective

Sliders

• Weirdness 0–100 %, 50 is "expected". Covers 10–30, experimental 70+.
• Style influence: vague tags want 70–90, very specific tags 40–60.
• Variety (v6): can rewrite your style tags; set 0 when you need them preserved.
• Change one slider per test. Max Mode helps acoustic/orchestral, does little for electronic.

Things that don't work well

• Adjective soup (epic, beautiful, cinematic): describe audible behavior instead.
• Prompt adherence decays after ~60 s: front-load the distinctive elements; build long songs with Extend or assemble sections in Studio.
• Lyric bleed: short poetic lines in the style box get sung. Keep that box dense and technical.
• DAW numbers (-14 LUFS, Q 1.4) do nothing in a prompt.