Suno Arrangement Prompts: Describing What Each Musician Does, Not What Genre It Is

Effective suno arrangement prompts describe players, not categories. That is the whole idea, and it is why the most common fix people reach for does not work.

The pattern reported over and over in Suno v6's first week goes like this. The intro arrives with a committed guitar riff and real drums. The vocal enters. The guitar thins to a pad, the drums simplify, and the verse becomes a bed for the singer. The natural response is to add more force to the style box — heavier, riff-driven, guitar-led, powerful — and the natural result is another track that does exactly the same thing.

Genre adjectives answer "what kind of music is this". They do not answer "what is the guitarist playing in bar nine". Arrangements are made of the second kind of information, and if you do not supply it, the model supplies it for you.

Why the verse is where it collapses

The thinning is not random and it is not a bug. It happens where the vocal enters, because the model is protecting intelligibility — pulling instruments back so the lyric sits on top. Any mixing engineer would do the same thing. The problem is that it does it more aggressively than most people want, and it does it by default unless told otherwise.

That gives you the actual instruction. You are not asking for the guitar to exist, which it already does. You are asking for it to continue underneath the vocal, which is a different request and needs saying in those terms. "Riff-driven" is a description of the song. "The guitar riff continues under the vocal through the verse" is a description of a decision.

Naming the section matters as much as naming the behaviour. An instruction with no location attaches itself wherever the model likes, which is usually the intro — where it was already going to play.

The section-by-section instrument map

This is more writing than a style tag, and it is the approach that survives past the first chorus.

For each section of the song, state which instruments are present and what each one is doing. A worked shape, for a rock track:

• Intro — electric guitar plays the main riff alone, drums enter on the second repeat
• Verse — guitar continues the riff at lower intensity underneath the vocal, bass follows the root, drums hold a steady backbeat
• Pre-chorus — drums build, second guitar layer enters
• Chorus — both guitars full, drums open up, bass drives eighth notes
• Bridge — guitars drop to a single sustained line, drums pull back to rim

Two things make this work. Every section is accounted for, so nothing is left to discretion. And every instrument is described by behaviour rather than by tone — "continues the riff at lower intensity" is an action, "crunchy guitar" is an adjective.

Sections you leave undescribed are sections the model fills on its own. That is fine when you do not care. It is the cause of the hollow verse when you do.

Solo instrument is not the same as instrumental

These two get conflated constantly, and the resulting confusion is entirely predictable once you see it written down.

Instrumental means no vocal. It does not mean sparse. Ask for an instrumental track and you will usually get the full arrangement with the singer removed, which is exactly what you asked for and rarely what people mean.

Solo instrument means one player and nothing else. If that is what you want, three things have to be true in the prompt: name the specific instrument, state that there are no other instruments, and exclude the usual suspects explicitly — drums, bass, pads, strings. "Solo guitar" on its own reads as a lead line over a band at least as often as it reads as one guitar alone.

This is a long-standing request rather than a v6 one, and most of the guidance circulating about it predates the current model. Exclusion behaviour has changed across generations before, so test it on v6 rather than trusting a workflow written for v5.

A worked rock example

Here is the difference written out, using the same song both ways.

The version that hollows out:

Heavy riff-driven rock, powerful guitars, energetic drums, anthemic chorus

Four adjectives and a genre. Every one of them describes the song as a finished object, and none of them tells any player what to do in any section. The model fills the gaps, and its default in the verse is to make room for the vocal.

The version that holds:

Intro: electric guitar plays the main riff alone, drums enter on the second repeat.

Verse: guitar keeps the riff going underneath the vocal at lower intensity, bass follows the root notes, drums hold a steady backbeat with closed hi-hat.

Pre-chorus: drums build, second guitar layer enters underneath.

Chorus: both guitars full, drums open to crash, bass drives eighth notes.

Bridge: guitars drop to one sustained line, drums pull back to rim clicks.

Longer, unquestionably. But every section is accounted for, every instrument is described by what it does, and the verse — the section that collapses — has the most explicit content of any of them.

Which fix for which symptom

Muffled output is two problems wearing one coat

"It sounds muffled" is one of the most common launch-week complaints and one of the least useful descriptions, because it covers two unrelated faults.

Masking. Too many instruments occupying the same frequency range, so the mix reads as dull even though every part is present. This is an arrangement problem and prompting fixes it — remove layers, thin the pads, give the vocal a clear range to sit in. A track with three guitars, strings and a synth pad will sound muddier than one with two of those, no matter what you write about clarity.

A dull render. The generated audio itself is soft, flat, or lacking high end, independent of what is in the arrangement. Prompting does not reach this. The routes are stems plus a mastering pass, covered in our Suno stems guide and AI music mastering pages, or simply re-rolling.

Telling them apart takes one test. Generate the same prompt several times. A fault that appears every time is in your prompt. A fault that appears on some rolls and not others is in the render, and you are re-rolling rather than rewriting.

If your arrangement tags are being rewritten

Before assuming a prompting failure, check whether your prompt is still your prompt when it reaches the model.

Suno's own v6 FAQ states that the Variety slider works by "adjusting and updating your style prompts", and that to retain full control of your style tags you should reduce Variety to 0. A carefully built instrument map can be partially rewritten by a non-zero Variety setting, which produces exactly the symptom people read as the model ignoring them.

Each of the creative controls does something different and they are easy to confuse. Suno style influence, Variety and Weirdness explained covers them one at a time against the documentation.

The gate that arrangement does not move

Once the arrangement works, the track faces a completely separate test, and it is worth being clear that the two have nothing to do with each other.

Distributor ingestion systems screen for the statistical fingerprint that AI generation leaves in the audio, plus the provenance data embedded in the file. They do not evaluate musicianship. v6 was built in partnership with Warner Music Group, BMG and Believe, and Suno has committed to audio watermarking and fingerprinting, so that trail is strengthening rather than weakening. In our benchmark on raw, unprocessed AI tracks: DistroKid rejected 50 of 50, TuneCore 47 of 50, CD Baby 42 of 50, against approximate thresholds of 0.78, 0.82 and 0.85.

Undetectr works on Suno v6 output and is the tool in our benchmark that addresses that layer. It operates on the audio signal across six layers — spectral artifact removal, temporal pattern normalisation, dynamic range processing, metadata sanitisation, and removal of embedded SynthID and C2PA watermarks. Because those are classes of artifact rather than a version-specific signature, a new model generation does not break it. Those are precisely the markers DistroKid, Spotify, Apple Music and Amazon Music scan for on ingestion.

Arrange it, clean it, then ship it. Our guide to uploading Suno tracks to Spotify covers the last step.

How we checked

The Variety slider behaviour is quoted from Suno's v6 FAQ, fetched 11 September 2026. The Weirdness and Style Influence definitions come from Suno's Creative Sliders documentation, fetched the same day.

Community reports come from a structured pull of Reddit and YouTube discussion covering 12 August to 11 September 2026: 25 Reddit threads carrying 782 upvotes and 1,351 comments, plus six YouTube videos with transcripts. v6 launched on 9 September, so direct v6 experience in that pool spans two days, and several of the instrument-control discussions predate the release entirely.

Three caveats we would apply to our own work. Guidance written for v5 exclusions may not hold on v6 and we have flagged where that risk applies rather than restating it as current. The section-map approach is a structural technique rather than a documented feature, so treat it as a starting shape to test against your material. And our distributor rejection figures describe a pre-v6 corpus; the v6 re-run is in progress and will be published when it completes.