Most Suno prompts describe a snapshot — a genre, a set of instruments, a mood that holds for three minutes. Songs don’t work that way. They move at two scales: microtiming, where notes sit against the grid, and arc, how the arrangement changes across sections. Those two scales are controlled from two different fields, which is why arc instructions written into the Style field usually do nothing.
The field split
The Style field describes the whole track. Anything you put there applies for the full duration. The Lyrics field carries bracketed metatags that attach to a specific point in the song.
| Scale | What it controls | Where it goes |
|---|---|---|
| Feel (microtiming) | How the players sit against the beat, for the whole track | Style field, plain descriptors |
| Arc (structure) | What changes, and at which section | Lyrics field, bracketed metatags |
This is the correction that matters most. “Chorus drops into half-time” in the Style field asks a whole-track description to carry a section-level event. The model has no anchor for when. Written as a metatag on its own line directly above the chorus, it has one.
Weak
Style field:
neo-soul, 88 BPM, sparse first verse,
builds one element per section,
everything drops out for the bridge,
chorus drops into half-timeFour structural events, no anchors. The model averages them into a vague dynamic wash or ignores them.
Working
Style field:
neo-soul, 88 BPM, swung hi-hats,
bass and drums behind the beatLyrics field:
[Verse 1]
[Sparse]
...
[Chorus]
[Half-Time]
...Scale one: feel
Microtiming is the strongest single lever against the gridded, mechanical quality that marks AI-generated music. It belongs in the Style field, roughly in position three to five — after genre and vocal description. Keep it to two or three descriptors on a vocal track; a crowded rhythm section steals room the vocal needs.
Behind the beat
Players land fractionally late. Relaxed, heavy, dragging — the feel underneath most great soul and hip-hop. The most reliable term in this article.
loose drumming slightly behind the beatAhead of the beat
Fractionally early. Urgent, pushing, slightly anxious. Punk, funk, anything that should feel like it’s accelerating without changing tempo.
hi-hats pushed slightly ahead of the beatSwing
Uneven subdivision — long-short rather than even-even. A notated, well-defined musical concept, which is why the model handles it reliably. Jazz, shuffle, boom-bap.
swung hi-hats, loose shuffle feelSyncopation
Emphasis on the offbeats. Also a precise musical term rather than a vibe, so it steers well. Name the instrument it applies to — applied globally it muddies everything.
syncopated bassline, emphasis off the downbeatPocket — weak alone
The zone where the rhythm section locks and the groove feels effortless. This is a description of a result rather than a technique, and on its own it’s closer to a compliment than an instruction. It appears to help in combination with a directional term, and does little by itself. Treat it as a modifier, not a control.
bass and drums locked in a deep pocket, slightly behind the beatScale two: arc
Arc terms go in the Lyrics field, each on its own line directly above the section it governs. Tags are case-insensitive and can be stacked on one line — [Chorus] [Half-Time] tells the model both things about the same section.
Sparse
Few elements, deliberate empty space. Works in both fields, and the difference matters: in the Style field it means the whole track stays thin; as a tag it means this section only.
[Verse 1]
[Sparse]Build
Accumulating energy and layers. Most effective placed on a pre-chorus or bridge, where there’s somewhere to arrive.
[Pre-Chorus]
[Build]Drop out
Sudden removal of elements. The most reliable way to make whatever follows feel large — contrast does the work, not volume.
[Bridge]
[Breakdown]
[Vocal Only]Low-passed as an arc device
Rolling off the high frequencies makes a section sound muffled and distant. As a texture note in the Style field it dulls the entire track. As a tag on an intro, it sets up the moment the filter opens.
[Intro]
[Low-Pass Filter]
[Chorus]
[Full Spectrum]Where the scales meet: half-time
Half-time is the one term that belongs to both. The drums feel half as fast while the tempo stays put — a feel change, but one that happens at a structural moment. That makes it a tag rather than a Style descriptor.
[Chorus]
[Half-Time]It’s also the cheapest weight in music. A chorus that drops to half-time sounds heavier than the same chorus played louder, and costs nothing in the mix.
What doesn’t work
Two habits appear constantly in prompt guides and neither survives contact with the model.
- Timestamps. “Drums enter at 0:30” assumes the model works to a clock. It doesn’t — generation length varies and there’s no reliable mapping from seconds to structure. Use a section tag instead: put the drums’ entry at the top of verse one.
- Stacking four structural events in the Style field. Each one competes with the others for a description that has no timeline. You get an average of all four, which usually reads as generic dynamics.
One honest caveat on everything above: this is ten terms, not forty. Feel and arc are a narrow slice of prompting, and most of the vocabulary circulating for them is either a texture term in disguise or a description of a result rather than an instruction. Ten that steer output beats forty that fill a page.
Worked examples
Both fields, filled together.
Neo-soul, dragging feel, half-time chorus
STYLE FIELD:
neo-soul, 88 BPM, Rhodes and upright bass, warm smoky female vocal,
swung hi-hats, bass and drums slightly behind the beat, tape saturation
LYRICS FIELD:
[Verse 1]
[Sparse]
(your lyrics)
[Pre-Chorus]
[Build]
(your lyrics)
[Chorus]
[Half-Time]
(your lyrics)
[Bridge]
[Breakdown]
[Vocal Only]
(your lyrics)Indie rock, pushing feel, live arc
STYLE FIELD:
indie rock, late 90s, 148 BPM, jangly Telecaster, raw untrained male vocal,
drums pushed slightly ahead of the beat, recorded live to tape, mono leaning
LYRICS FIELD:
[Intro]
[Guitar Only]
[Verse 1]
(your lyrics)
[Chorus]
[Full Band]
(your lyrics)
[Bridge]
[Breakdown]
(your lyrics)
[Final Chorus]
[Build]
(your lyrics)
[Outro]
[Abrupt End]Minimal electronic, filtered opening
STYLE FIELD:
minimal electronic, 92 BPM, deep sub bass with long decay, muted plucked synth,
whispered female vocal, syncopated bassline, restrained throughout
LYRICS FIELD:
[Intro]
[Low-Pass Filter]
[Verse 1]
[Sparse]
(your lyrics)
[Drop]
[Full Spectrum]
(your lyrics)
[Bridge]
[Half-Time]
(your lyrics)Diagnostics
| Symptom | Cause | Fix |
|---|---|---|
| Track is rhythmically stiff | No microtiming language at all | Add a directional term to the Style field — behind or ahead of the beat |
| Structural instructions ignored | Written in the Style field, which has no timeline | Move them to the Lyrics field as tags above the relevant section |
| Every section sounds the same | No arc tags anywhere | Tag at minimum one sparse section and one full one — contrast needs both ends |
| Dynamics feel vague and mushy | Several structural events competing in the Style field | Keep the Style field to whole-track feel; one event per section tag |
| “Pocket” did nothing | Describes a result, not a technique | Pair it with a directional term or drop it |
| Timing changed but so did tempo | Half-time read as a tempo instruction | Keep BPM in the Style field and half-time as a section tag — they’re separate |
| Whole track sounds muffled | Low-passed used as a Style descriptor | Move it to an intro tag so it has something to open into |
Questions
Yes. The Lyrics field accepts structure tags with no lyrics under them, so you can lay out an entire instrumental arrangement as a sequence of tagged sections.
One structural instruction per section is the reliable pattern, stacked on one line where two things apply to the same section. Guidance suggests front-loading matters — tags in the first 20 to 30 words carry the most weight. Tagging every section heavily tends to flatten the contrast you were trying to create.
Training data is weighted heavily toward verse-chorus-verse with a climactic final chorus, and that shape reasserts itself unless something pushes against it. Explicit section tags are that push. For anything meant to stay level — background music, loops, ambient — say so directly rather than assuming the absence of build instructions will produce a flat arrangement.
The vocabulary does — behind the beat, swing, half-time and syncopation are standard musical terms, not Suno syntax. The two-field split and the bracketed tag format are Suno-specific.
Next step
Take a generation you already like and regenerate it with one change: move every structural instruction out of the Style field and into the Lyrics field as section tags. Nothing else. That single move demonstrates the field split more clearly than any amount of vocabulary.