Song Breakdown Loop
Investigate a reference song, break it into usable musical and lyrical parts, and record it in your research library.
Anyone who says "make it like that song" — you want your AI to actually understand what makes the song work and save that, so the study compounds instead of repeating.
When someone references a song, they usually mean two things at once: put it in that song's SUB-GENRE, and give it ASPECTS like that song. This loop pulls both apart so the new song lands in the right sound — not a clone, a grounded direction.
It breaks the reference into helpful, reusable bits: the musicality (BPM, key, instruments, energy, mood, structure, vocal style, production) and — when the ask is to WRITE it that way, not just sound that way — the lyrical style. (Exact-text songs care less about the words; based-on songs use the writing style heavily.)
It works from real signals — reviews, descriptions, any analysis of the song's unique aspects — and is honest about what's verified vs inferred. Every finding is saved to your private research library so you (and the AI) can pull it up next time.
Any sub-genre it turns up that isn't in the genre tree yet gets added, so the taxonomy grows with your research.
Skills & actions it uses
The concrete, reusable skills this loop calls to actually do the work.
Song investigationPull a reference song apart into musicality + lyrical style from reviews, descriptions, and analysis — verified, not guessed.agentResearchUpsert / List / GetRecord the breakdown in the owner's private research library, and recall what's already been studied so nothing is re-researched.agentEnsureGenreNodeAdd any sub-genre the song belongs to that isn't in the tree yet.
The loop
- 1
Identify the song and its sub-genre
Pin down the song (and the specific version/arranger if it matters), then the genre-tree node it sits in. If that sub-genre isn't in the tree yet, add it.
- First check the research library — maybe it's already studied
- Note public-domain status; it changes what later loops may use
- 2
Break down the musicality
From reviews, descriptions, and analysis, capture the concrete musical signals.
- BPM / tempo, key / modal color, time feel
- Instrumentation and production texture
- Energy, mood/emotion, and the song's structure / arc
- Vocal style and dynamics
- 3
Read the intent — sound vs written
Decide whether they want it to SOUND like that song (musicality only) or be WRITTEN like it (lyrical style too). Capture the lyrical style only when 'written-like' is in play.
- 4
Check the lyrics (for based-on)
For non-exact-text work, study how the song is written — themes, point of view, devices, rhyme/cadence — as inspiration. (Skip for exact-text songs; the words there come verbatim from the source.)
- 5
Record it
Save the entry to the research library: title, year, genres + sub-genres, the musicality breakdown, lyrical style, a ready style prompt, notable versions, sources, PD status — reusable on every future song.
Then run it again for the next song, chapter, or track.
You end with
- A recorded song-research entry in your private library — musicality + lyrical style, source-backed
- A grounded style direction (sub-genre + aspects) for the new song
- Any new sub-genre added to the genre tree
Skip this loop if…
- You want a one-off vibe you'll never reuse — recording it is the whole point here.
- You want to clone the actual recording — this studies a song to evoke it, never to copy it.
How it connects
Pairs with
Commonly run just before or after this one in the pipeline.
When someone says "make it like that song," study the song instead of copying it — turn it into a grounded style direction and a reusable artist entry.
Research an artist deeply once, store them in your Artist Library, then channel that voice on every future song — so the study compounds instead of repeating.
Study the seed, brainstorm wide, lock a voice, then write a performance-ready lyric and style prompt.