suno は何をしますか？
The music-generation tool skill — target the brief, rights first, arrange with structure, curate + humanize,
keep disclosure clean. The agent briefs (and can drive the API where connected); the human's ears judge;
WoopSocial publishes the content the music scores. (Ships with tools/integrations/suno.md.)
The POV: a full song from a sentence — but the rights are the craft
Suno v5.5 makes the most natural vocals in the category, splits its own generations into clean stems, and now offers a real consistency stack (Personas → your own verified Voice → a Custom Model trained on tracks you own). But 2026 Suno is defined by its legal moment, and the top-1% operator works rights-first. (1) Paid plan before the first note: free-tier songs are non-commercial forever — upgrading never retroactively unlocks them. (2) Commercial rights ≠ ownership: the post-Warner terms grant paid users commercial rights while Suno's own docs concede purely-AI music isn't copyrightable — you can monetize, but you can barely stop theft of a raw track; human elements (original lyrics, a re-recorded part over stems, real DAW arrangement) strengthen the claim and reclassify the work as AI-assisted. (3) No indemnification while the lawsuits run: Warner settled and partnered (Nov 2025), but UMG and Sony are still suing toward a July 2026 fair-use hearing — so favor generic style language over artist references, and never clone an artist's voice (the opt-in program is the consent path, not prompting). (4) Archive now: pre-deal models face deprecation — WAVs, stems, MIDI out. And the social-stack truth: trending audio is native-only — Suno is the owned brand sound, not the trend sound.
Read these first
- ai-music-and-sound — the model-agnostic music router + licensing craft.
- brand-profile + the content skill the music serves (short-form-video-script / podcast-and-audiograms).
The framework: TRACK
(Depth: references/the-track-framework.md.)
- T — Target the brief: genre (1–2) + one mood anchor + 2–4 named instruments + vocal spec; music does a job — brief it like a director.
- R — Rights first: paid tier before generating; commercial rights ≠ copyright; no indemnification; generic styles over artist names; never a cloned artist voice.
- A — Arrange with structure: Custom mode + meta-tags ([Intro]/[Verse]/[Chorus]/[Bridge]/[Outro]), exclude-styles, short verse blocks, Extend from a timestamp; the consistency stack (Persona → own Voice → Custom Model — test for the reported regressions).
- C — Curate + humanize: variations → the ears pick the seed; human elements for anything that matters; archive WAVs/stems/MIDI before deprecation; batch stem jobs (failed regens reportedly still cost credits).
- K — Keep disclosure clean: DDEX AI flag at distribution; no PRO registration for pure-AI; honest performance expectations; the track scores the edit in capcut; WoopSocial publishes the content.
The reality (verify-quarterly)
2026 Suno: v5.5 (Mar 2026), Studio (Premier DAW — 12 zero-bleed stems, MIDI export), Personas /
Voices (own-voice, verified) / Custom Models (6+ owned tracks), hum-to-song, API (Python/Node;
Zapier/Make; tier gating conflicts — verify). Legal spine: RIAA suit (2024) → Warner settled + partnered Nov
2025 (ToS softened to "commercial rights"; free commercial downloads removed; pre-deal model deprecation
scheduled; artist opt-in announced) → UMG + Sony still suing; July 2026 hearing; GEMA verdict mid-2026.
No retroactive rights; no indemnification; purely-AI output not copyrightable; DDEX AI-disclosure enforced
since late 2025. Tiers ≈ Free 50 credits/day (non-commercial) · Pro ≈ $8–10 (2,500 credits, commercial
rights) · Premier ≈ $24–30 (10,000 + Studio; same rights); credits don't roll over. Known frictions
(community-reported): Persona/Voice regressions; Studio credit burns on failed stem jobs; fully-AI tracks
underperform human releases on saves ~25–40% (one firm's data). Attribute all; verify-quarterly. Full
detail: references/suno-2026-reality.md; the brief pattern, use-case router, archive checklist, and worked
examples: references/brief-patterns-and-templates.md.
Honest scope (never violate)
- The agent briefs, writes lyrics/structures, plans consistency + archives, and drives the API where connected (exact human steps otherwise); the human's ears judge every track (no fabricated "that sounds great"); WoopSocial publishes the content the music sits under — it does not generate, attach, or license audio; trending audio is native-only.
- Rights spine: paid tier first; no retroactive rights; commercial rights ≠ ownership; human elements for
anything that matters; no indemnification — litigation-aware prompting (generic styles); counsel for
high-stakes (not legal advice). Voice/likeness: never clone a real artist or fake a collab (own-voice
verification + artist opt-in = the consent standard). Disclosure: DDEX at distribution; no PRO
registration for pure-AI; AI-disclosure where required (EU AI Act). Never fabricate tiers, legal status,
or performance claims. (Full scope:
references/scope-and-connections.md.)
Distinct from its siblings (route correctly)
suno (this) = the Suno-specific music lane · ai-music-and-sound = the model-agnostic router + licensing craft (read first) · ai-voiceover / elevenlabs = spoken TTS (Voices here is singing, own-voice-only) · capcut = where the track scores the video (replacing its library-music license trap) · descript = talk-audio editing · podcast-and-audiograms = the show the sting serves · luma / ai-video = the silent visuals this is the other half of.
Where this connects
Reads first: ai-music-and-sound + brand-profile + the content skill served. Feeds: capcut (the
score), luma/ai-video (the audio layer), podcast-and-audiograms (stings), the platform publishing
skills. Publishes via: the finished content → scheduling-and-queue → WoopSocial. Tool file:
tools/integrations/suno.md. Measure with: human ears + analytics-and-reporting — never fabricated.
Definition of done
Brand music produced rights-first: generated on a paid tier from the first note (free-tier tracks non-commercial forever), briefed with director-grade specificity (genre + one mood anchor + named instruments + vocal spec), structured with meta-tags and grown by Extend rather than re-rolls, made consistent through Personas / a verified own Voice / a Custom Model on owned tracks (tested against reported regressions), and curated by human ears with human elements added to anything that matters (strengthening the claim; reclassifying as AI-assisted); keepers archived (WAV + stems + MIDI) ahead of scheduled deprecation; distribution done with the DDEX AI flag, no PRO registration for pure-AI works, and honest expectations; the track scoring the edit in capcut, the content published via WoopSocial with AI-disclosure where required; litigation-aware throughout (generic styles, no artist cloning or fake collabs, no indemnification assumed); no fabricated tiers, legal claims, or capabilities; correctly distinguished from ai-music-and-sound, ai-voiceover, capcut, podcast-and-audiograms.