Decide when to stop generating before the search becomes the song

Set the search budget before the first output

Generation can produce enough variation to postpone every musical decision. Set a bounded budget: a number of prompts, credits, minutes, or rounds; a fixed brief; and a date when one candidate must move into an editable session. This is not a rule against exploration. It is a guardrail that protects arranging, performing, editing, and listening time from being consumed by the next possible take.

Write success criteria that the ear can test: a four-bar opening with usable negative space, an energy lift at a specified section, no named-artist imitation request, and a harmony that can support a human vocal. “Surprise me” may be a valid exploratory instruction, but it is not yet a selection rule.

Separate generation from selection

Label outputs by prompt and attempt, then make a first pass without editing them. Select a small number for a second listen against the brief. If a take passes, name exactly what survives: groove, contour, texture, section contrast, or sound palette. Move that idea into a DAW or notation workspace before generating more. This differs from auditioning a generator, which compares products; here the tool is fixed and the decision is when a project has enough raw material.

Use explicit stop rules

• Stop when one candidate meets every must-have criterion and no failure blocks editing.
• Stop when the next output is unlikely to answer a new question.
• Stop when repeated outputs reveal that the brief, not the sample count, needs revision.
• Pause when rights, export, or account terms are unclear for the intended use.

A stop can be “select and arrange,” “rewrite the brief,” or “change tools.” It does not have to be “release this audio.”

Worked decision log

A producer allows 12 20-second instrumental sketches for a 16-bar intro. The brief requires a sparse opening, pulse by bar five, and room for a spoken line. After eight attempts, take 08 has the right pulse but an overfull opening; take 11 has a useful opening but no lift. The log says: keep take 11’s texture and take 08’s rhythmic idea; stop at 12; rebuild the intro with owned instruments. The decision is not that either output is finished. It is that the search has delivered its two useful inputs.

Limitations

Budgets do not guarantee originality, quality, rights clearance, or a satisfying arrangement. Some commissions require broader exploration, and a real deadline may change the threshold. Keep prompt, model, plan, and output records for material you retain. Then use the pre-prompt brief and editing an assisted performance to turn the chosen idea into music.