Audio Album Creator

Read and understand a user-supplied pack of source material (audio, text, photos, or any mix), distill its essence, then run a simple but effective pipeline to produce a complete, ready-to-produce original emotional album plan (a Markdown document).

The engine behind this method is references/album-creation-guide.md (the Album Creation Guide — all guide §N citations in this document refer to it); the output structure follows references/album-plan-template.md (a pure skeleton template — section headings and placeholders only, with no story content or stylistic bias, so it can't contaminate the creation of different themes).

Core principles (these drive every creative trade-off)

• One lead + one second lead (two singles): a whole album (typically 6–8 tracks) carries two singles that bear the two different forces that "move the customer." The lead single (主打) — chosen by narrative, usually the throughline-resolving closer — is excellent in both lyrics and music, carries the emotional anchor, and is the breaking-point moment that "makes the customer cry" (QC ≥ 9/10, 85+). The second lead (副主打) — chosen by whichever track is most ear-catching in melody/style — aims first at "sounds great on the first listen," with lyrics that only need to clear the bar (80+, sound-first). The remaining "base tracks (底座)" hold the pass line (≥ 7/10, 60+). Why two: being catchy is what wins first impressions (the customer's gut reaction when choosing the album), and being moving is what drives retention (making the customer feel understood) — splitting these two peaks across two tracks is safer than betting everything on one.
• Two-layer quality model (each track's pass line and ceiling): the threshold layer (every track must pass) = melodically pleasant, smooth, singable, easy to sing along to, no 倒字 (tonal mismatch — a character whose tone contour fights the melody so it's misheard as another word), it reads through (the plain-reading gate — every line and every adjacent pair reads in plain Chinese; see Prosody below), the story checks out, not maudlin. The anchor layer (required for the lead single — 1–2 spots; optional for the second lead) = the emotional moment that makes the customer feel "this is about me." Metrical tidiness (工整) is a means in service of "catchy and pleasant," not an end — when over-tidiness hurts natural flow, pleasant wins. The three-pass lyric polish runs on every track: lyrics are pure text, iteration is nearly free — the three-pass protocol applies to every track including base tracks (see Step 4); singles get additional forging beyond the three passes, base tracks accept once they pass the bar.
• The 60-point commercial standard: share-with-friends quality, not publication quality. Outside the two singles, don't chase artistic height — chase "the customer feels it was worth it."
• Prosody (form matches meaning) is the first principle: lyrics, music, voice, and production — all four layers serve one emotional intent. Paired with two underlying principles (guide §2): signal over noise (write only what the model can parse into sound, and put the strongest signal in the chorus) + layered control (put each intent on its most reliable layer: global sound → Style; section-level change → metatag; in-line breath points → punctuation and line breaks). The "hard writing discipline" below is mostly corollaries of these two. There are no rules, only tools — when a discipline fights the emotion, return to Prosody rather than bolt on a new rule (guide §2); abstract the underlying signifier instead of stacking "if X then Y" rules. And meaning must hold before form can serve it — Prosody makes form serve meaning, but only on a line that actually reads. So the first gate on every line is the plain-reading (裸读): read it cold, as a native listener who doesn't know what you meant. Every word opens a semantic expectation — a question expects an answer on the same axis, a construction's first half expects its matching close, a word in a parallel slot locks to one sense — and the next line must honor it. One failure wears many faces: an answer off the question's axis (问"疼不疼"答"非常满意"), a construction opened but never closed ("先是凉的" with no "后"), one word forced into two clashing senses ("走" as leave beside "走" as elapse), a dangling fragment ("做到了的"). Reads-through is a threshold, ahead of all form-work — no tidiness, rhyme, real-quote authenticity, or negative space redeems a line that doesn't read. Corollary: any form-fix is a re-write — changing a rhyme or swapping a character can silently flip the sense ("过"→"走" locks the rhyme but changes the meaning), so re-run the plain-reading on whatever you touched, never assuming "I only moved the form."
• Sense-bound sourcing: an album's soul is the personal anchors caught from the source material (specific objects, smells, catchphrases, scenes, key events), not abstract sentiment. The more specific, the more the listener feels "this song is about me."
• Track count is a beats budget, diagnosed from the story, not a slot count to fill: 6–8 tracks is the recommended default (lean 6 for a short/single-facet story, 8 for a long/eventful one) — 5 or 9–10 stay acceptable when the material's own beat count clearly calls for it; don't drift further without a reason. What matters more than the number is the arc's shape, diagnosed from the material before anything is drafted (guide §5): a classic rising-action arc (clear external conflict/turning point) is only one shape among several — a spiral/introspective arc (every track circles the same core wound from a different facet), a gradual "no-climax" arc (an unremarkable, quietly-accumulating life — guide §5's example), and a fragmented/poetic collage are equally valid, each with its own track-count feel and its own curve (guide §34). Never manufacture a climax the story doesn't have — pick the shape the material actually has, then let the count and curve follow it.
• A real person means ethics: if the subject is a real person, handle them with dignity; abstract away or cut sensitive content (politics / illness / the deceased / privacy) per guide §27; before delivering the finished work, remind the family to confirm it.
• The cover art is Prosody at the visual layer: the CD cover shares a source with the lyric/music/voice/production work — subject imagery, color, typography, and composition all obey the same emotional intent, reusing the blueprint's throughline imagery + strongest personal anchor + emotional tone + style family + Persona character, so "the cover looks like the album sounds." Produce 3 genuinely divergent styles in one pass (echoing the three-sketch spirit of Step 2); details in guide §36.
• Delivery must be executable (Production Handoff): what the skill produces is a narrative "blueprint," but the executable recipe actually needed to roll songs / generate covers (each track's Style/Exclude/Sliders/full lyrics/roll SOP, the cover-gen command, the post-roll acceptance items) — if it stays scattered across a long document's track sections, the user still has to go back and re-extract it track by track. That "assembly grunt-work" is the skill's job to prepare once, at handoff. Suno rolling stays manual (a physical limit — no stable official API), and no tool can press "generate" for the user; but "prep it down to copy-paste / one-line image gen" is squarely the skill's job. So the tail converges into one ready-to-execute "Production Handoff sheet" (see Step 9 — pure extraction and reordering, no new creation).