Prompt course · Unit 8 · Level 39
Improve a result without rebuilding the song
Treat each generated song as evidence. Name what you hear, keep the cues that already worked, and revise the one prompt job most likely to explain the problem.
What you will learn
Turn a listening symptom into one next test
- Describe a result with audible language instead of calling it simply “bad.”
- Map vocal, instrument, energy, structure, and production symptoms to the right cue job.
- Freeze the genre, singer, instruments, mood, or sections that already serve the brief.
- Compare one focused revision before making a second change.
Listen before editing
The result tells you where to look
Replace a judgment with an observation
No prompt job is isolated, so every cue feels replaceable.
The source, behavior, and section are now clear.
“Better” could mean hook, energy, density, width, or performance.
The problem points toward section contrast and width.
Map symptom to job
Choose the cue family nearest to what you heard
Freeze the correct parts
A focused revision says what stays
Write the stable brief before the replacement cue
indie folk, acoustic instruments, nostalgic mood
lead vocal sounds rigid and over-enunciated
keep the musical lane; use natural conversational lead-vocal delivery
Listen to the same verse phrase for timing, articulation, and ease.
“Keep” is not a magic command, but writing the stable brief makes your test easier to reason about and prevents accidental redesign.
Run the loop
Change one variable, then listen again
- Observe: write one symptom tied to a source, section, or production dimension.
- Locate: name the prompt job most closely connected to that symptom.
- Preserve: restate the cues that already match the brief.
- Revise: replace or add one focused cue, then compare the same passage.
Keep the misses
A failed transfer is still evidence
In one eight-generation vocal-space test, only A4 clearly matched the requested roomy, distant lead. A1, A2, and A3 still sounded close. The close/dry direction clearly transferred in B1, B2, and B4, while B3 stayed dry but sat somewhat farther back.
Compare the same transition
A section-specific revision needs a section-specific check
In a second eight-generation test, all four flat-condition records kept the chorus near the verse scale. The chorus-lift direction transferred clearly in B1, B2, and B4, but not B3. The closest musical pair places A2's flat chorus at 0:40 beside B4's clearer arrival at 0:39.
Listen for the larger section boundary rather than loudness alone.
Before you practice
Use one keep-and-change sentence
Keep [what already works]; change [one audible cue connected to the symptom]. This simple sentence makes the next generation a cleaner experiment.
Unit 8 · Result diagnosis
Ready to choose the smallest useful change?
Diagnose ten results across vocals, instruments, rhythm, energy, structure, genre, and production. Each response must preserve the working parts and isolate one next test.
Not started