A music generation request does not jump straight from words to sound. It passes through a sequence of stages, each producing something the next stage consumes.
The first stage interprets the prompt. A sentence such as "warm lo-fi hip hop with a mellow piano loop" is not itself music; it is a description that must be translated into musical intentions — a genre, a mood, a rough tempo, a set of instruments, and a target length. This translation is what makes the rest of the process possible, because the later stages need concrete targets rather than adjectives.
The second stage plans structure. Before any detailed material is produced, the system decides the overall shape of the piece: how many sections there will be, how long each lasts, and how they connect. A three-minute track might be planned as an intro, two verses, a chorus, and an outro, each with a duration. Planning at this level keeps the piece from drifting, because the generator knows what it is filling in.
The third stage generates the musical material. Working within the planned structure, the system produces the actual content — melody lines, chords, bass, and rhythm — section by section. This is where the approach families from the previous chapter do their work, but the important point here is that generation is guided by the plan rather than starting from nothing.
The fourth stage renders to audio. If the material exists as symbolic data — notes, chords, and timing — it is still silent. A rendering or synthesis stage converts that symbolic representation into an audio waveform that can be played. Only at this point does the result become a listenable track.
The fifth stage is review and refinement. Users listen to the output, compare alternatives, edit sections, and re-prompt. The first result is treated as a draft rather than a final product, and the loop between listening and adjusting is part of how the system is normally used.
Ordering these stages matters because each one constrains the next. A vague prompt limits what the structure planner can do; a weak plan limits how coherent the generated material can be; and no rendering means no sound at all.