Suno Song Builder

Turn a user brief, genre archetype, or reference into an original, production-aware Suno v6 prompt or full song package. Treat Suno as a probabilistic composition and production engine—never as a uniqueness, copyright, or symbolic-MIDI guarantee.

1. Request Routing Decision Tree

Identify the primary deliverable and load only the required references:

Infer harmless defaults (2:40–3:20 full song; 2:30–3:15 film cue with a 40–60s hero passage; check the live Duration slider for the current maximum) and ask at most one concise question only when a missing choice materially alters the work (e.g., vocal vs. instrumental when unresolved).

2. Deterministic Song Engineering Protocol

Step 1: Write the Creative Thesis & Originality Delta (or Opt-In Parody Map)

1. One-Sentence Thesis: A [function] about [concrete physical situation/decision], moving from [emotional state A] to [state B], identified by [one audible relational signature rule].
2. Originality Enforcement vs. Opt-In Parody (references/songcraft.md):
   • Default (Original Songs & Reference Translation): Extract only abstract functional appeal (energy curve, pocket, ensemble relationship, mix space). Never include an artist name, producer name, song title, copied lyric, or recognizable progression in any field. Change at least 5 identity-bearing dimensions (at least 7 for famous/iconic sources): premise/title, hook phonetic shape, melodic contour, harmonic engine, groove/meter, section order, lead/counter-voice roles, signature sound event, recording space, and emotional resolution.
   • Opt-In Parody / Lyric Adaptation (only when explicitly requested): Map the source skeleton first (syllables per line, stressed-beat positions, rhyme scheme, and open vowel families on held notes), lock new stressed syllables to the original beats, use monosyllabic swaps at hook anchors, and keep Style 100% free of artist or song-title names.
   • Multi-Song Batches: Vary all 6 anchors (Premise, Hook shape, Groove, Harmony, Lead identity, Form) across every option.

Step 2: Engineer the Style Field (Mustango arXiv:2311.08355 & MusicRL arXiv:2402.04229)

Captions are ambiguous by default: MusicRL reports that one caption maps to very different music for different users. Keep Style 100% positive (move all negations to Exclude) and within the live character counter (third-party sources report 1,000 characters, unconfirmed by Suno; obey any tighter user cap). Select the mode that fits the request:

• Mode A — 6-Stage Studio Signal Chain (best for fast style prompts, genre locks, and the 83 sonic archetypes in references/studio-signal-chain-and-archetypes.md): [01. Era & Subgenre], [02. Exact BPM (68–174BPM)], [03. Core Timbres (specific amps, synths, samplers, acoustic models)], [04. Vocal Delivery (register, mic distance, texture)], [05. Scene & Mood (emotional arc & cinematic setting)], [06. Mix & Master (reverb space, console/tape saturation, stereo imaging)] Example: Dreamy 80s synth-pop, 110BPM, vintage Juno-106 poly-synths, shimmering 16th-note arpeggios, ethereal female vocal, hazy nostalgic midnight mood, spacious Lexicon hall reverb, warm analog chorus modulation
• Mode B — 8-Part Relational Studio Architecture (best for picture scoring, reference originality translation, or custom multi-state arrangements): [1. Function & target duration] -> [2. Primary genre lane + 1 contrast] -> [3. BPM/tempo feel, meter, & pocket] -> [4. Harmonic engine/mode] -> [5. Focal ensemble roles (2–3 core hardware/acoustic instruments + counter-voice)] -> [6. Vocal register, mic distance, & delivery (or instrumental lead)] -> [7. Section energy arc across >=3 states including subtraction] -> [8. Recording space, console/reverb finish, & 1 signature rule]
  • Focal Ensemble Rule (LeVo arXiv:2506.07520): Anchor around 2–3 primary instruments with specific hardware/acoustic timbres and relational verbs (muted Telecaster answers the vocal cadence; upright bass holds a pedal drone) rather than a 10-instrument noun list.

Step 3: Engineer Lyrics or Instrumental Structure (SongComposer arXiv:2402.17645 & YuE arXiv:2503.08638)

• Prosodic Stress, Rhyme Blend, & Vowel Placement: Keep Lyrics within the live character counter (third-party sources report 5,000 characters). Match stressed-beat counts across parallel verse lines without forcing awkward word inversions, blend perfect rhyme (for closure/hooks) with family/slant/internal rhyme (for momentum and tension), and place open sustainable vowels (ah, oh, ay, oo) on long hook notes.
• AI Vocalist Phonetic & Pronunciation Hygiene (references/songcraft.md):
  • Spell numbers and symbols out as sung words (twenty four seven, nineteen ninety-nine).
  • Space or hyphenate acronyms (A-I, N-Y-C) unless sung as a word; respell or hyphenate tricky homographs/proper nouns (thru, bi-o-en-gi-neer-ing).
  • Use restrained delivery punctuation only where needed: hyphenated vowel extension (lo-o-o-ove) on melismas, ellipses (I... need... you) for breath pauses, and ALL CAPS sparingly for belted peaks.
• Concrete Evidence Over AI Clichés: Use tangible objects, places, actions, and interpersonal stakes; ban default AI filler (neon shadows, echoes in the dark, ashes rising, scars, chains, destiny) unless literally required by the brief.
• Clean Section Tags (YuE arXiv:2503.08638): Place concise structural tags ([Cold Open | dry single voice], [Verse 1], [Pre-Chorus], [Chorus | drums drop out], [Bridge], [Outro]) alone on their own lines with blank lines between sections. Never stack contradictory metatags or put sung text on the same line as a [Tag].
• Instrumental Cues: Enable Instrumental and structure by dramatic state (Establish Motif $\rightarrow$ Build Motion $\rightarrow$ Subtraction/Drop $\rightarrow$ Hero Passage $\rightarrow$ Edit Tail/Resolve).

Step 4: Populate Exclude & Calibrate Suno v6 Controls (JASCO arXiv:2406.10970)

• Exclude Field: Comma-separated list of unwanted audible traits only (e.g., stock risers, trap hi-hat rolls, supersaws, four-chord pop loop, heavy autotune, stadium reverb, over-compressed master).
• Suno v6 Model & Mode Selection (references/suno-controls.md):
  • v6 (Flagship — Pro/Premier): Default for polished, reliable, genre-accurate production.
  • v6-wild (Experimental — Pro/Premier): Recommend when the user wants bold, unpredictable textures or avant-garde motif discovery before converging in core v6.
  • v6-mini (Fast/All Tiers): Recommend for free-tier users or rapid lyric/meter/pronunciation sketching.
  • Variety slider (0 vs. above 0): Set Variety to 0 to keep full control of style tags during controlled A/B takes; raise it for wider variation.
  • Max Mode (extra credits; Suno gives no multiplier): Recommend for songs longer than 2 minutes, covers that stay close to the original, style transfer, and consistent vocals/style through the track.
• Slider Heuristics (references/controllable-song-research.md): Clearly label numeric slider suggestions as practical heuristics (e.g., Exploration: Weirdness 45–65% / Style Influence 50–70%; Convergence: Weirdness 20–40% / Style Influence 75–90%; Owned Audio Seed / Mashup: Audio Influence 75–90%, avoiding hard 100% artifact transfer).

3. Tiered Copy-Ready Delivery Schema

Lead immediately with the requested deliverable, keep wrapper prose short, and place every paste-ready field in its own standalone fenced text block:

Tier 1: Fast Style / Sonic Archetype Requests

When the user asks for a quick Style string, sonic archetype, or backing vibe without lyrics:

1. Style:

   [6-Stage Studio Signal Chain prompt string]

2. Exclude (include when helpful to protect the sonic pocket):

   [Comma-separated unwanted traits]

Tier 2: Complete Suno v6 Song Packages

When the user asks to build a song, score, or full prompt package, deliver in this order:

1. Creative direction: 1 concise sentence/paragraph stating the thesis, signature rule, and (when adapting a reference) the originality delta.
2. Title:

   [Original title]

3. Style:

   [Positive-only Mode A (6-Stage Studio Signal Chain) or Mode B (8-Part Relational Brief)]

4. Lyrics (or Instrumental structure):

   [Section-tagged, prosody- and phonetics-checked lyrics, or dramatic instrumental cue map]

5. Exclude:

   [Comma-separated unwanted instruments, styles, vocal traits, and mix habits]

6. Settings: Compact summary of Model (v6, v6-wild, or v6-mini), Mode (Custom), Instrumental toggle, Duration target, Max Mode (On [extra credits] vs. Off), Variety (0 vs. above 0), and heuristic slider ranges (Weirdness, Style Influence, Audio Influence).
7. Iteration (keep to 1–2 concise bullets by default; expand into full SongEval arXiv:2505.10793 diagnostics, v6 natural-language edit commands, Song Editor section replacement, Suno Studio Auto-Stretch / cent-level pitch, 12-stem export, and Remaster steps when troubleshooting or requested).

4. Verification Gate & Diagnostic Repair

Before delivery, verify (and run scripts/prompt_lint.py when validating files):

• Zero Artist/Producer/Title Leakage: No living artist names, producer names, reference titles, or copied hooks appear in Title, Style, Lyrics, or Exclude.
• Length & Positive Style Hygiene: Style and Lyrics fit the live counters (third-party-reported 1,000 / 5,000 characters when the UI is not visible), Style contains zero inline negations, and at least 3 distinct energy states (or a clear dynamic arc) are planned.
• Prosody, Phonetics & Tag Hygiene: Section tags sit alone on their own lines; numbers and acronyms are formatted for natural vocal pronunciation; verses use concrete physical imagery and natural spoken stress.
• Evidence Honesty: Settings distinguish official Suno v6 controls (v6/v6-wild/v6-mini, Max Mode, Variety, sliders) from practical numeric heuristics; un-auditioned audio is never described as heard or verified.

When repairing an existing Suno v6 render (references/songcraft.md), isolate the failure layer (composition, arrangement, prosody/pronunciation, timbre, mix, or local section glitch), preserve the winning seed/section, change one variable per A/B batch (with Variety: 0), and use v6 natural-language editing (e.g., "Change the chorus arrangement to stripped upright bass and dry close harmony" or "Change the lyric in Verse 2 from X to Y") or Song Editor section replacement before regenerating the entire track.

Do Not

• Do not upload or sample third-party music, film audio, or another person's voice in Suno (including in v6 mashups or multimodal prompts), or promise copyrightability, uniqueness, or detector evasion.
• Do not bury negative exclusions inside Style or Lyrics instead of using Exclude.
• Do not reskin the same chord loop and form across a multi-song batch.
• Do not purchase credits, change plans, enable Max Mode (extra credits) without user consent, or publish songs publicly when operating the live Suno UI.

5. Resource Index

• references/studio-signal-chain-and-archetypes.md: The 6-Stage Studio Signal Chain, Tempo Pill bands (68–174 BPM), moderation-safe hardware/amp/synth/mic/console/reverb lexicon, and 83 engineered copy-ready sonic archetypes across 6 suites.
• references/songcraft.md: Intake defaults, 9-dimension reference appeal map, 10-point originality delta, opt-in parody & lyric adaptation skeleton mapping, thesis & musical engine design, prosody/rhyme/phonetic hygiene, 6-anchor batch diversity, and diagnostic repair table.
• references/suno-controls.md: Audited Suno v6 controls snapshot (2026-10-30), v6 / v6-wild / v6-mini model selection, Max Mode, Variety, Duration/Weirdness/Style/Audio sliders, natural-language editing, multimodal & mashup rules, Custom Models, Voices, Suno Studio (Clip Settings, cent-level pitch, Auto-Stretch), 12-stem separation, live UI rules, and official links.
• references/controllable-song-research.md: Controllable music generation research (Mustango arXiv:2311.08355, JASCO arXiv:2406.10970, SongComposer arXiv:2402.17645, MusicRL arXiv:2402.04229, YuE arXiv:2503.08638, LeVo arXiv:2506.07520, SongEval arXiv:2505.10793, NegBench arXiv:2501.09425) with paper findings separated from craft heuristics, plus v6 mode/slider starting ranges.
• scripts/prompt_lint.py: Standalone CLI linter and self-test suite checking Style length, inline negations, artist/producer name leaks, standalone [Section] tag formatting, and unexpanded digit strings in Lyrics.
• scripts/validate_coverage.py: Repository validator checking required files, frontmatter keys, routed links, all 83 numbered archetypes (01–83) across 68–174 BPM, and zero artist-name leakage.