Jarvie Music Writing Loop
Study the seed, brainstorm wide, lock a voice, then write a performance-ready lyric and style prompt.
Anyone using an AI agent to write original, replay-worthy songs from a seed (a theme, a story, a scripture, a journal entry) who wants real craft instead of generic AI slop.
This is the writing engine that sits at the heart of every song you make. You hand it a seed - a feeling, a scenario, a passage of text - and it does the work a real songwriter does: it studies the idea from many angles, brainstorms far more material than it needs, deliberately chooses a genre and an artist voice to write inside, and only then writes a lyric built to be performed, not just read. The output is a finished spec: a title, a structured lyric with embedded performance cues, and a pasteable style prompt.
The loop is built to fight the single biggest failure mode in AI songwriting: the 'concept developer' who writes thoughtful mini-sermons with choruses nobody remembers. Instead it forces convergence on one killer hook and one main image, and it pushes every verse to be the listener's actual life rather than a recap of the source text. The research and brainstorm work is real and can run long - for a scripture seed it drops into a study mode and reads the passage from many angles, pulling lines that can be reused directly or lightly rewritten to sing better. Save those research and brainstorm artifacts as documents attached to the song so the thinking is never thrown away.
Crucially, this loop nests inside a platform create loop - on Suno, the Suno create loop - which renders your finished spec into actual audio. This loop's job is to hand that create loop something worth rendering: a lyric and style prompt so well-chosen that two takes from one submission give you something you would replay. The genre and artist decisions are where originality is won or lost, so the loop treats them as deliberate choices - especially when your seed only named a broad genre family and there is room to play with sub-genres and mixtures.
Skills & actions it uses
The concrete, reusable skills this loop calls to actually do the work.
suno-songwritingThe craft engine itself - mindset, hook design, structure, artist-voice translation, and prompt economics that this loop runs on.scripture-songwritingFires when the seed is a sacred text - drives the study-mode research, theme selection, and POV/voice choices for scripture seeds.suno-addThe handoff into the create loop - submits the finished title, lyric, and style prompt to your chosen workspace to render two takes.
The loop
- 1
Quick read and creative brief
Open by studying the seed and writing a short brief. Summarize the seed in 4-6 bullets and lock the narrative spine before any lyric exists. Save the brief as a document on the song so later steps and re-runs build on the same foundation.
- Name the core feeling, the scenario, and the stakes in plain words
- Capture the subtle details from the seed that will become texture, not narration
- Decide POV (I / you / we), tense, and the emotional arc: where it starts, where it shifts, where it lands
- If the seed didn't specify a style, default to one that matches the idea you're trying to portray
- 2
Research the seed from many angles
Drop into study mode and write the thinking out - this can run many pages and lives as a research document on the song. The goal is to surface more raw material and more honest emotion than a single read would give you.
- Read the seed from multiple angles: whose POV, what power dynamic, what it demands of the hearer, what the central image is
- For a scripture or text seed, study the passage closely and pull quotable lines and images straight from the source
- Find the modern mirror: translate each source feeling, dynamic, image, and demand into a present-day equivalent
- Write it all down - the study artifact is the well you draw verses and hooks from later
- 3
Brainstorm wide - diverge before you converge
Generate far more candidate material than you'll use, then save it as a brainstorm document on the song. Quantity here is what kills AI slop later - you choose from a real field instead of settling for the first idea.
- 8-12 title options - short, singable, hook-adjacent
- 6-10 hook lines, roughly 6-10 syllables, chantable, image-loaded
- Several candidate choruses built around the strongest hooks
- Source lines pulled directly from the seed, plus lightly-rewritten versions that sing faster or cleaner
- Lean on hook patterns that work: command + image, confession + image, reframe, counter-thesis, negation, question-that-becomes-answer
- 4
Converge - pick the title and hooks with tests
Narrow the wide field down to one title and 2-3 primary hooks using explicit tests, so the choice is earned rather than arbitrary.
- Stranger test: can someone who doesn't know your seed instantly get it and sing it back after one listen?
- Chorus test: can the whole chorus revolve around this hook without clutter?
- Repetition test: does it still sound good chanted six or more times?
- If a hook fails a test, drop it - don't force it
- 5
Decide genre, sub-genre(s), and artist voices
Run this in parallel with brainstorming. Choose deliberately - this is where originality is won. Propose a few distinct directions, then lock one and say why it serves the heart of the seed.
- Propose 3 distinct style directions, each with genre/era/texture/tempo/vocal attitude
- When the seed only named a broad genre family, choose specific sub-genre(s) and mixtures - there's room to play
- For each direction, name 2-3 inspiration artists as study references only (in or out of the genre, chosen as strong writers)
- Lock ONE direction and one voice, and write a sentence on why it fulfills the request
- Keep artist names in your thinking only - the final platform prompt describes the voice with adjectives and craft language, never by naming a living artist
- 6
Structure the song so contrast is obvious
Outline the sections with intent before writing lines, so each part has a distinct job and the song doesn't flatline. Pick a structure that fits the chosen style's pacing.
- Verse = story details and image; Pre-chorus = tension and lean-in; Chorus = the thesis and payoff; Bridge = a twist, decision, or new lens
- Default arc: Intro, Verse 1, Pre, Chorus, Verse 2, Chorus, Bridge, Final Chorus, Outro
- Target 2:30-3:30 for most songs; stretch to 4:00 only for anthems and finales
- Account for how the chosen style sings - a faster or denser style needs fewer or shorter lines per section
- 7
Write the lyric for performance
Write the actual lyric optimized for a singer and a generator, not for the page. Embed performance and instruction cues, and stay aware of length and pacing as you write. The generator over-focuses on lyrics, so instrumental or no-singing sections must be instructed explicitly.
- Verse 1 is the LISTENER'S life with source imagery as texture only - never a recap of the seed
- Verses conversational and image-rich (or one strong analogy); 'don't say it, feel it' - let the image carry the meaning
- Pre-chorus holds the breath: shorter lines, a lift; chorus clean, quotable, minimal clutter
- Use [bracket] cues on their own lines for sections and production direction; use (parentheses) only for sung ad-libs and call-response
- Decide instruction density on purpose — both valid: HEAVY (tag [instrumental break]s, solos, register flips, no-vocal moments — required or they won't happen) or LIGHT (few tags, let the engine's style defaults arrange it)
- Write for the singer's mouth: respell hard proper nouns phonetically (Ephraim → 'Ee-fraim', Moroni → 'Moe-roh-nye') and expand archaic contractions (e'en → even, o'er → over, 'tis → it is) so they're sung right
- Keep the channeled artist in your head only — never name a living artist in the lyric or the final style prompt
- Trim or expand lines so the lyric actually fits the song length the style implies
- 8
Build the style prompt from sub-genre + artist elements
Translate the locked sub-genre and artist voice into a tight, pasteable style description shaped for the target platform's Style box. This is the bridge into the create loop.
- Shape: [Genre/sub-genre] then 3-5 instruments, then vocal type, then one emotional anchor, then a duration target
- Front-load the genre - platforms weight the first words most heavily
- Describe the artist voice with adjectives and craft language only - never name a living artist in the final prompt
- Respect the platform limit (on Suno the Style field is ~200 characters) and don't repeat what's already in the lyric tags
- State the vocal mode explicitly: rap vs sung vs spoken vs a mix
- 9
Hand off to the create loop and triage
This writing loop nests inside a platform create loop. Pass the finished spec - title, performance lyric, style prompt - to the create step, render, then listen and decide.
- Call suno-add to submit the title, lyric, and style prompt to your chosen workspace
- Each submission renders two takes - listen to both before deciding
- Keep, iterate, or scrap: if the hook hits but verses are weak, rewrite verses; if the lyric hits but the sound is wrong, change the style prompt; if both are mediocre, the spec is the problem, not the render
- Don't iterate a fundamentally weak concept more than a couple of times - pivot to a new angle or a different artist voice instead
Then run it again for the next song, chapter, or track.
You end with
- A creative brief and a long research/study document on the seed, saved as documents on the song
- A wide brainstorm artifact: 8-12 titles, 6-10 hooks, candidate choruses, and reusable/rewritten source lines
- A locked creative decision: genre + sub-genre(s) + one artist voice, with the reason it fits
- A performance-ready lyric with embedded [cues], sung (ad-libs), and explicit instrumental/no-vocal instructions
- A pasteable, platform-shaped style prompt built from the sub-genre and artist elements (no artist names)
Skip this loop if…
- You just want a one-click sketch and don't care whether the lyric is original or memorable
- You want the AI to imitate a specific living artist's identity rather than write an original voice
- You have no rendering platform - this loop produces a spec that a create loop turns into audio
How it connects
Alternatives
A competing loop for the same job — different tool, taste, or depth. Pick whichever fits you.
Pairs with
Commonly run just before or after this one in the pipeline.