research node active
00:00:00 ESC: back

Directional Over Timing: Add / Remove / Hold

Core finding: the model responds better to structural direction than timestamps. You steer the arrangement like a conductor, not a stopwatch.

featured promptlab metatags suno workflow

What we learned

  • Directional cues (“thin out”, “bring in”, “hold”, “suppress”) beat literal timestamps.
  • Instrument suppression is as important as instrument addition.
  • Thinning before impact makes later arrivals more likely to register.
  • Blank-line silence is a valid pacing tool: it “resets” attention.

Practical rule set

  • Write instructions like you’re guiding a mix engineer.
  • Use “REMOVE/HOLD” to create space before “ADD”.
  • Call out one instrument as the anchor per section (bass OR drums OR lead), not everything at once.
  • Use short lines. Avoid huge paragraph prompts that lock the model into inertia.

Example architecture (copy template)

[SECTION — DIRECTIONAL ARRANGEMENT]
[HOLD: drums (tight, minimal)]
[REMOVE: lead synth]
[KEEP: bass (present, steady)]
[THIN: mids (leave air)]
[SUPPRESS: sax, vocals (delayed induction)]

[BUILD]
[ADD: percussion (ghost hits)]
[ADD: texture layer (tape flutter / noise halo)]
[HOLD: kick (do not overdrive)]

[IMPACT]
[ADD: lead instrument (single identity)]
[WIDEN: stereo image]
[RESTORE: mids]
[SUSTAIN: groove]

Why it works

The model treats the lyrics box like a high-level “scene plan”. When you give it a timeline, it often ignores it. When you give it a sequence of intent, it behaves like it’s following a mix script.

Next iteration

  • Turn these directives into a reusable “Prompt Lab Skeleton” file.
  • Add a small vocabulary list of reliable verbs: hold, suppress, thin, restore, induce.
  • Test if “delayed induction” improves instrument persistence across rerolls.