Arrangement: give every instrument one job

Crowding is a role problem

When a generated track sounds cluttered, the instinct is to blame the mix. Usually the mix is fine and the arrangement isn't: you asked for six instruments and gave none of them a defined job, so several ended up doing the same one in the same frequency range.

Two rhythm guitars, a piano and a pad all playing sustained chords in the same octave will sound like mud no matter how well they're balanced. Nothing is wrong with the audio. The parts are simply redundant.

The diagnostic question is simple: if two instruments were playing the same notes in the same octave, would you be able to tell them apart? If not, one of them is redundant, and adding it made the track denser without making it bigger. Density and size are different things, and confusing them is the root of most cluttered arrangements.

Four roles cover most arrangements

Every part in a working arrangement is doing one of four jobs. Naming the job in the prompt is more effective than naming more instruments, because it tells the model what the part is for rather than just what it sounds like.

Write roles into the prompt

Note what the last line does. Telling the model explicitly to keep a range clear works better than listing fewer instruments, because you keep the richness while removing the collision. The phrase 'answers in the gaps' is doing similar work — it defines when the trumpet plays, not just that it exists.

The same technique works in reverse when something is missing. If a track feels thin, don't add an instrument — give an existing one a second job, or specify a register nobody is occupying. 'Bass moves into a melodic line under the chorus' adds perceived size without adding another voice to argue with.

Foundation: upright bass (roots, walking), brushed kit (soft pulse)
Harmony: Rhodes (mid-range comping, sparse)
Lead: vocal; muted trumpet answers in the gaps
Colour: vibraphone accents on section changes only
Keep the mid-range uncluttered; vocal sits forward

Dynamics by section

The other half of arrangement is what drops out. A chorus feels big mostly because the verse before it was smaller, and generated tracks often sound flat because everything plays all the time.

• Verse: foundation plus one harmony instrument. Leave space.
• Pre-chorus: add one element or lift the existing one; hint at what's coming.
• Chorus: full arrangement, add colour, widen.
• Second verse: pull back but not all the way — keep one chorus element.
• Bridge: remove something significant; contrast is the point.
• Final chorus: everything, plus one element held back until now.

Diagnosing a crowded track

• Can you hear the bass line clearly? If not, something else is in its range.
• Count the instruments sustaining chords — more than two is usually the problem.
• Is there more than one melodic line in the same register at the same time?
• Does anything ever stop playing? If not, add a drop.
• Is the vocal competing with a pad or a guitar in the same octave?
• Could you name every instrument on one listen? If not, one of them isn't earning its place.

Fewer parts, better defined

There's a useful reference point here: most of the records you love have fewer instruments than you think. Count the parts on a song you consider rich and full, and the number is usually four to six, each clearly audible. The impression of fullness comes from arrangement and space, not from part count.

The reliable fix is almost always to remove an instrument and give the survivors clearer jobs. A four-piece arrangement with distinct roles sounds bigger than an eight-piece one where everything overlaps, because the listener can actually resolve the parts.

Test it the same way as everything else: remove one instrument, generate twice, listen. If you don't miss it, it wasn't doing anything. That's a slightly uncomfortable test to run on a part you liked writing, which is exactly why it's worth running.

FAQ