Controlling Song Structure
Sections need bar counts and entry points; naming verse and chorus alone specifies nothing.
Related
A section name is not a section
Write verse, chorus, bridge and the model often returns four minutes of one loop that never exhales. A section name is only a label until it is tied to a bar count, an energy level and instruments entering or leaving. Real structural information is: how long this part runs, who enters first, who drops out, and which chord it lands on.
Write the sections as a table
| Section | Bars | Entry | Exit | What to hear |
|---|---|---|---|---|
| Intro | 8 | Solo piano | Four-beat drum fill | Does the theme appear early |
| Verse | 16 | Bass and drums together | Vocal leaves space at phrase ends | Any change every four bars |
| Chorus | 16 | Full band, open hi-hat | Everything stops on the last beat | Does the energy actually lift |
| Bridge | 8 | Drums out, harmony shifts colour | Rising fill back to the chorus | Does it add new information |
| Outro | 8 | Lead restates the theme | Fade or clean stop | Does it end on the tonic |
The exit is the column people skip. A clear entry with a vague exit glues two sections together. One beat of silence at the end of a chorus pushes the next part away better than any adjective.
Say who moves first at the transition
A transition is not a natural flow but a specific action: a drum fill, the bass landing half a bar early, the harmony shifting to the minor subdominant, everyone resting for a beat. “A sixteenth-note snare fill in the two bars before the chorus, then a beat of silence” is executable; “build the emotion” is not. Connecting sections with one motive sounds smoothest, while a key or metre change needs preparation before the seam.
Duration and section count pull against each other
Five sections in three minutes leaves thirty-six seconds each, and the bridge gets four bars, too short to develop. Cut a section or make the bridge a four-bar interlude instead of forcing everything in. Editing after generation is usually safer than getting it right in one pass, but cutting removes tails and reverb, so crossfade briefly at the seam and confirm the downbeat has not slipped half a beat. When a piece has more than six sections and runs under four minutes, question the structure before adding more prompt text.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- Walter Piston, Harmony 与 William Caplin, Classical Form（Oxford）中关于段落划分与终止式的判断。
- 文本提示里的段落标记与结构控制，依据公开模型文档与制作实践。