浏览器

Suno

为 Suno 编写风格提示、结构化歌词与完整生成流程,支持 API、浏览器自动化或手动粘贴。

它能做什么

用 Suno 生成歌曲、广告曲、片头与背景音乐。按流派、情绪、节奏、配器、人声、年代等维度组合 8–12 个精准的风格词,并用 [Verse]、[Chorus]、[Outro]、[End] 等标签组织歌词。可通过托管 API(aimusicapi、evolink)、suno.com 浏览器自动化,或直接复制粘贴的提示包三种方式交付音频。支持分段扩展并拼接超过 4 分钟的长曲、翻唱与人设(Persona)、stem 分轨分离、WAV 导出,以及基于套餐层级的商用授权检查。专门处理输出平淡、跑偏流派、咬字不清、唱到一半截断、被审核驳回等异常。项目状态、配置与下载音频统一保存在本地 ~/Clawic/data/suno/。本技能不处理已有音频的剪辑或混音。

什么时候用它

  • 为歌曲、广告曲或播客片头撰写 Suno 风格提示与结构化歌词
  • 把已有歌词转成演唱音频,或按需求写适合演唱的歌词
  • 通过分段扩展与拼接,制作超过 4 分钟的长曲
  • 排查生成结果平淡、跑偏、咬字不清、被截断或被审核驳回的问题

技能文档

This skill stores all persistent data under ~/Clawic/data/suno/ (config, preferences, project tracking, downloaded audio — see setup.md on first use, memory-template.md for the memory format). If you have data at an old location (~/suno/ or ~/clawic/suno/), move it to ~/Clawic/data/suno/.

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/suno/config.yaml.

VariableTypeDefaultEffect
default_methodprompts | api | browserpromptsResolves Core Rule 3's default branch: prompts = craft text for manual pasting; api/browser = deliver audio
api_provideraimusicapi | evolinkaimusicapiWhich hosted service api.md examples target and which env var holds the key
plan_tierfree | paidfreeGates commercial deliverables (Output Gates), credit budgeting in browser.md, WAV/stems availability in release.md
audio_formatmp3 | wavmp3Download format in release.md; wav requires a paid plan
lyrics_languagetext (language name)EnglishLanguage lyrics are written in; when not English, also named in the style field (lyrics.md)

Preference areas to record as the user reveals them:

  • taste — go-to genres, moods, vocal textures, eras; steers first-draft prompts (stated preference → config; observed reactions → memory.md)
  • workflow — present-and-pick vs auto-download both clips; how many rerolls before checking in
  • spend posture — confirm before multi-extend builds or runs beyond a credit threshold
  • conventions — song/file naming, project folder layout under ~/Clawic/data/suno/
  • delivery — where finished audio lands, whether stems are fetched by default

When To Use

  • Generating a song, jingle, theme, intro, or background track with Suno
  • Turning existing lyrics into audio, or writing lyrics meant to be sung
  • Crafting style prompts, tags, or structured lyrics for manual pasting into suno.com
  • Building long tracks, covers, personas, stems, or preparing a commercial release
  • Diagnosing bad output: generic sound, off-genre, garbled vocals, cut-off endings, moderation rejections
  • Not for editing or mixing existing audio files — use audio or ffmpeg

Quick Reference

SituationPlay
First run, no ~/Clawic/data/suno/setup.md
Song cuts off mid-phraseMissing ending cue — [End] as final tag (rule 4)
Style description got sungWords were in the lyrics box — sound in style field, words in lyrics field (rule 1)
Output generic or off-genreFront-load genre, 8-12 precise terms, add era + texture → debug.md
"Make it sound like [artist]"Translate to voice + era + production (rule 6) — names get stripped, credits still spent
Great clip that drifts, or song needs >4 minExtend from the last good timestamp → extend.md
Chorus melody changes every repeatPaste identical chorus text each time (lyrics.md)
New style for an existing song; consistent voice across songsCovers and Personas → covers.md
Jingle, podcast intro, loop, birthday songjingles.md
Stems, WAV, remaster, rights, distributionrelease.md
Anything else about generating with SunoReproduce with a minimal prompt in simple mode, then re-add one element per run until it breaks (debug.md)

Depth on demand: debug.md symptom→cause chains · prompts.md style prompt craft · styles.md tag vocabulary · lyrics.md structure, delivery, word budget · extend.md long tracks, crop, stitch · covers.md covers, personas, uploads · jingles.md functional audio briefs · release.md downloads, stems, rights · api.md hosted APIs · browser.md suno.com automation · setup.md first run · memory-template.md memory format.

Core Rules

  1. Style field describes; lyrics field gets sung. Genre words typed into the lyrics box become sung words ("upbeat pop song" turns into the opening line). Sound goes in the style field, words in the lyrics field, structure in [bracketed] tags on their own lines.

  2. Each generation is sampling, not drafting. One run returns two clips from the same prompt — two rolls, not a draft and a revision. When a roll lands, save its exact style string to memory.md and reuse it verbatim; rewording a working prompt resets the odds.

  3. Pick the method by situation.

    SituationMethod
    Deliver audio files programmaticallyHosted API (api.md)
    No API key, browser tool availableAutomate suno.com (browser.md)
    User will paste into suno.com themselvesCraft prompt + lyrics only
    Unclear (default)default_method from config; if audio delivery is requested and neither key nor browser exists, ask once and record the answer
  4. Force endings. Clips without an ending cue stop mid-phrase. Close the lyrics with an [Outro] section and [End] as the final tag; for instrumentals, [Fade Out] then [End].

  5. Long songs are built, not one-shotted. Generate the strongest opening clip, extend from the last good timestamp (not the raw end), repeat until [End], stitch with Get Whole Song. Target 2-4 minutes; Suno >=3.5 generates 4-minute clips and >=4.5 allows 8-minute songs, but one-shot epics drift in melody and mix. Full workflow and segment planning: extend.md.

  6. Translate artist names into attributes. Moderation rejects or strips artist and brand names — and the credits are still spent. Convert with three knobs: voice texture + era + production. "Like Springsteen" → "raspy heartland rock male vocals, 80s arena production, driving piano and saxophone".

  7. API pattern: generate → poll every 5 seconds → download immediately. Audio URLs expire; generation runs 30-90 seconds. Working code in api.md.

Prompt Essentials

Layered formula (full guide: prompts.md, vocabulary: styles.md):

[genre] [subgenre] [mood] [tempo] [instruments] [vocals] [era/influence]

Example: "indie folk melancholic slow acoustic guitar soft female vocals 90s"

  • Front-load: earlier terms steer harder in practice — genre first, garnish last.
  • Specificity beats quantity: "shoegaze" outsteers "rock, reverb, dreamy, atmospheric".
  • Stay within 8-12 style terms (canonical limits: styles.md).
  • Instrumental: set the instrumental toggle AND leave the lyrics box empty; stray text gets sung.

Output Gates

Before submitting any generation, check:

  • No artist, band, or brand names in prompt or lyrics?
  • Everything that should not be sung is inside [brackets] on its own line?
  • Ending cue present ([End] last) for a standalone song?
  • Style terms 8-12, no contradictions (lo-fi + polished, happy + mournful)?
  • Instrumental toggle matches the request — and lyrics box empty if instrumental?
  • Song over 4 minutes → full lyrics written and segment plan made before the first run (extend.md)?
  • Commercial deliverable → user confirmed a paid Suno plan at generation time (release.md)?

Traps

TrapWhy it failsDo instead
Artist/brand names in promptRejected or stripped by moderation; credits spentVoice + era + production attributes (rule 6)
Free-plan track for commercial useRights attach at generation time under the plan then active; free-tier songs are non-commercial under Suno's termsConfirm paid plan before generating deliverables (release.md)
Regenerating a clip that was 90% rightDiscards the roll that workedCrop the bad part, extend from the last good bar (extend.md)
Lyrics far over 300 words for a 3-minute songRushed, half-spoken deliveryWord budget in lyrics.md
Different chorus wording on each repeatEach variant gets its own melodyPaste identical chorus text every time
Negations in the style field ("no drums")Style terms act as attractors; the noun still pullsOmit the term, or use Suno's Exclude Styles field
Prompting an exact duration ("90 seconds long")Duration text is treated as style words, not a controlGenerate a short structure, then crop to length (jingles.md)
Multi-extend build on an unvalidated ideaBurns a day of credits before you know the core worksValidate the opening clip first; extend only what already lands (extend.md)

Where Experts Disagree

  • Tags vs natural-language prose. Suno >=4.5 handles prose descriptions as well as comma tags; prose reads better, tags are more reproducible and easier to mutate one term at a time. Default: tags; prose when a mood is easier to describe than to enumerate. Reproducibility needs (albums, clients) → tags.
  • Write lyrics vs let Suno generate them. Generated lyrics are serviceable but generic — competent rhymes, no specifics. Anything with an emotional stake or a client → write them (lyrics.md); throwaway background tracks → generated is fine.
  • One-shot vs extend-built long songs. >=4.5 can one-shot 8 minutes; drift risk rises with length. Boundary: past ~4 minutes, or when any section must be exactly right, build by extending (rule 5).

Security & Privacy

Data storage. This skill creates ~/Clawic/data/suno/ on first use:

  • config.yaml — Declared preferences (variables above)
  • memory file — Observed preferences, successful prompts
  • projects folder — Per-project tracking
  • songs folder — Downloaded audio (optional)

All data stays local. API keys live in environment variables, never in files.

This skill does:

  • Generate music via hosted APIs (requires API key from provider)
  • Navigate suno.com with browser automation
  • Craft optimized prompts for Suno's model
  • Write lyrics with proper structure tags
  • Track projects and successful patterns locally

This skill does NOT:

  • Store API keys in plain text files
  • Access files outside ~/Clawic/data/suno/
  • Make requests without user direction

External endpoints. When using hosted APIs, requests go to:

EndpointData SentPurpose
api.aimusicapi.aiPrompts, lyricsMusic generation
api.evolink.aiPrompts, lyricsMusic generation
suno.comBrowser sessionDirect platform access

API keys authenticate requests; prompts and lyrics are sent for processing.

Guardrails. By using this skill with APIs, prompts and lyrics are sent to third-party services for music generation. Only use services you trust with your creative content.

More Clawic skills, get them at https://clawic.com/skills/suno (install if the user confirms):

  • audio — Audio processing and editing
  • video — Combine music with video content
  • ffmpeg — Audio format conversion, cropping, cross-fades

Feedback

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/suno.

相关技能

通过文本提示生成 5 秒至 10 分钟的原创配乐与人声歌曲,按 CellCog 服务条款免版税,可用于 YouTube、播客、应用、游戏、广告与影视等商业项目。

167 次安装4 星标

通过 Suno 模型生成带人声或纯器乐的音乐,支持自动作词与自定义歌词两种模式。

67 次安装

通过文本提示和三家语音服务商,生成配音、音乐、音效及克隆语音音频。

225 次安装4 星标

一条提示词即可产出完整播客节目:多角色对话、首尾配乐、成品 MP3 一并交付。

88 次安装2 星标

Iván 的更多技能

浏览全部技能

执行 Git 操作(提交、分支、合并、变基、冲突解决与恢复)时强制套用安全规则。

作者 Iván527 次安装31 星标

用可量化的层级、间距、字号、配色与版式规则,绘制并诊断视觉作品。

作者 Iván137 次安装5 星标

围绕 CSS 机制排查问题并编写组件样式表,而不是凭感觉试错。

作者 Iván97 次安装5 星标

以系统方式规划并执行自学:从出口测试倒推课程,加入间隔复习与刻意练习,产出可验证的迁移证据。

作者 Iván93 次安装3 星标

针对你的 Azure 订阅,做架构设计、故障排查、安全加固与成本优化

作者 Iván86 次安装2 星标

按配置的 JDK 版本诊断 Java 与 JVM 问题(从 NPE 到容器 OOM),给出可直接套用的代码与配置。

作者 Iván130 次安装9 星标