Music Mv Studio

音乐 MV（音乐视频 / 歌词可视化）全流程制作工作流。用于「给某首歌做 MV」「按歌词做分镜并生成画面」「把逐镜片段合成带字幕的成片」，也用于复盘优化已有的 MV 流水线。核心是先锁设计再生成、一切结论用像素说话：导演五层与开工必答 → 首帧闸门 → 逐镜生成（方向锁/结构锁/运动锁）→ 像素级验收 → 定点返修。附 73 条血泪清单（真实踩过的坑）与可运行脚本（scripts/）；Windows / macOS / Linux 三平台都有 CI 冒烟。

5 stars 1 forksImages, Video & Audio

Install

skills CLI (Claude Code, Cursor, Codex and more)

npx skills add chenqianshinian/music-mv-studio --skill music-mv-studio

Manual install

git clone --depth 1 https://github.com/chenqianshinian/music-mv-studio.git /tmp/music-mv-studio
mkdir -p ~/.claude/skills
cp -r /tmp/music-mv-studio/. ~/.claude/skills/music-mv-studio

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# Music MV Studio v3.4 · 音乐 MV 制作工作流

**这是什么**：一套把「一首歌 + 一份歌词」变成竖版 MV 的可复用工作流。设计层是导演方法
（叙事四问、导演五层、镜头语言纪律），执行层是一条已经跑过多首长片（每首约 100 镜）的管线
（关键帧闸门 → i2v 逐镜生成 → 按帧号缓存的合成 → 像素级质检 → 定点返修）。

**这不是什么**：

- 不是一个「输入歌名就自动出片」的黑盒。它要求你（或模型）先把**设计**做对——
  全片最大的返工来源不是模型不会画，而是**设计没定**和**验收靠眼睛**。
- 不包含任何素材：没有音频、没有成片、没有真人肖像。示例是虚构的 `song-a`。
- 不绑定某一家模型。默认实现走一个兼容 OpenAI 风格的接口（图像/视频/视觉各一个模型名，
  见 §11），换供应商只需要改 `.env` 里的模型名与地址。

**最值钱的部分不是脚本，是 §5 那 73 条血泪清单**：能跑的流水线很多，
把「踩过什么坑、为什么、正确做法是什么」写下来的很少。

## 目录

| 节 | 内容 | 什么时候看 |
| --- | --- | --- |
| §0 | 开工前必答 | **每首歌开工第一件事** |
| §1 | 核心原则 | 开工前读一遍 |
| §2 | 导演设计五层 | 写分镜前 |
| §3 | 执行管线与工具 | 开工时搭骨架 |
| §4 | 首帧闸门（不合格分五类） | 首帧生成完 |
| §5 | 血泪清单 73 条 | 出事时按现象查 |
| §6 | 提示词纪律（各种"锁"） | 写提示词时 |
| §7 | 字幕纪律 | 做字幕时 |
| §8 | 交付验收清单 | 交付前逐条过 |
| §9 | 成本与限流 | 排期时 |
| §10 | 已被推翻的结论（决策留痕） | 想改管线时先看这里 |
| §11 | 配置与环境变量 | 第一次跑 |
| §12 | 发布门禁与贡献 | 想提 PR 时 |

## 0. 开工前必答（**每首歌重新回答，不许沿用上一首的默认值**）

| # | 必答项 | 为什么不能默认 |
| --- | --- | --- |
| 1 | **风格媒介** | 写实／日系动漫／沙画／手绘／厚涂油画……每首歌最优解不同；观众在意"系列观感一致"，**换风格要主动报备** |
| 2 | **主唱性别与视角** | 决定"摄像机是谁的眼睛"。**问用户或直接听**，不要从人声分轨的基频反推（§5 #6） |
| 3 | **主角与叙事** | 谁、几个人、什么关系、关系怎么变；时间线（一天/一年/一生）；空间线（起点-途经-终点） |
| 4 | **镜头规模** | 总镜数（经验值：主歌 2.5–2.8s/镜、副歌 3.3–3.6s/镜）；先确认总时长与可接受的生成时间 |
| 5 | **字幕字体** | **只用 OFL 等可商用授权**（志莽行书、思源黑体、马善政）；禁止华文行楷/微软雅黑/方正系（§5 #5） |
| 6 | **字幕配色规则** | 语义色／情绪色／天空色……按歌定，不复用上一首（做法见 §7） |
| 7 | **要不要额外图层** | 例如"算法界面层""行情网格"。**永远不要让生成模型画界面或写字**，用代码画（§5 #25） |

## 1. 核心原则（违反必返工）

Read the full skill on GitHub

Details

Skill name: music-mv-studio
Repository: chenqianshinian/music-mv-studio
License: MIT
Repo updated: 8 October 2026

FAQ

What does the Music Mv Studio skill do?

音乐 MV（音乐视频 / 歌词可视化）全流程制作工作流。用于「给某首歌做 MV」「按歌词做分镜并生成画面」「把逐镜片段合成带字幕的成片」，也用于复盘优化已有的 MV 流水线。核心是**先锁设计再生成、一切结论用像素说话**：导演五层与开工必答 → 首帧闸门 → 逐镜生成（方向锁/结构锁/运动锁）→ 像素级验收 → 定点返修。附 73 条血泪清单（真实踩过的坑）与可运行脚本（scripts/）；Windows / macOS / Linux 三平台都有 CI 冒烟。

How do I install the Music Mv Studio skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add chenqianshinian/music-mv-studio --skill music-mv-studio. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Music Mv Studio popular?

It lives in chenqianshinian/music-mv-studio, which has 5 GitHub stars and 1 forks.

Related skills

Screenshot Infographic Skill

Wanqiu12345/screenshot-infographic-skill

把「一张截图 + 一段功能描述」或「一段主题文案」自动变成一整套可在小红书/公众号/社群直接发的风格统一、视觉优美的图。两种模式：①截图模式——生成功能教程图/产品导览图（1张概览+单功能细节，局部放大，品牌色自动跟随截图）；②纯文案模式——表意型排版信息图，纯排版零插画、离线可跑，视觉元素直接编码内容。晚秋简约风：主色 #1A3A6B 深蓝 + 哑光金 #B08D45 + 暖白 #FBFAF8。适合产品教程、功能安利、新手引导、知识科普、活动预告、资讯解读。触发词：截图生成教程、功能讲解图、产品导览图、截图配图、功能安利图、把截图做成教程、文案生成信息图、小红书配图、名词科普配图、infographic generator、screenshot tutorial、 feature guide image、text to infographic。

Images, Video & Audio

Moodle Mcq

danielcregg/moodle-mcq

Generate, review, and improve Moodle quiz questions. Supports GIFT format (compact, ~6x fewer tokens) and Moodle XML format (full features with syntax highlighting). Four modes: create (easy/standard/challenging difficulty), review (improve existing MCQs). Handles MCQ, true/false, short answer, matching, numerical, and cloze questions. Enforces answer length balance (15/15/70), self-contained stems, plausible distractors, and proper Moodle XML structure including the critical <name> tag fix.

Images, Video & Audio

Ugc Creator

0xAnni/ugc-creator

Turns Claude into a hyper-realistic AI UGC influencer studio. Use this whenever the user wants to create, manage, or scale a synthetic/AI influencer or AI UGC creator: building a consistent character, writing image or video generation prompts for an AI model (Flux, Midjourney, Kling, Sora, Veo, etc.), producing TikTok/Reels/Shorts-style content, product/testimonial/UGC clips, or "shot on iPhone" lifestyle photos of a recurring synthetic person. Trigger it even when the user doesn't say "skill": e.g. "make my AI model look the same in every post", "generate a week of UGC for my AI girl", "why d

Images, Video & Audio

Game Install Size Optimization

dungnotnull/game-install-size-optimization-agent-skill

Game Install Size Optimization harness - production-grade asset pipeline & install-size analysis. Use when optimizing game install size, reducing download/storage, implementing compression/streaming/dedup, or analyzing game asset distribution. Triggers on: game install size, asset compression, download optimization, storage reduction, texture/audio compression, asset streaming, language packs, delta patches, SteamPipe, game distribution.

Images, Video & Audio

Xxd Panel 046

nevertoday/xxd-panel-046

XXD Panel 046｜Impasto Micro-World: rebuilds the photo's subject as a small, tactile 3D miniature painted in thick impasto oil, standing on one clear diagonal band of thick paint — water, ground, a path or a strip of light — across bright white textured paper with wide whitespace, with reflections, ripples or light spots where they fit and only a few impasto clouds, a sun or mist behind. Visible paint ridges and palette-knife marks; bright, fresh colours (sky blue, lake teal, fresh green, warm yellow, coral, peach) drawn from the photo. A short title with a little small text. Outputs PNG top-bo

Images, Video & Audio

Firsttake

emersonsklar/firsttake

Use when planning, prepping, running, or debriefing a video shoot where the user is responsible for the recording — camera and audio settings, lighting placement, on-set checklists, footage QC, verified card offload and archiving, or a delivery spec for a post house. Also use for post-production on any footage, however it was shot — syncing angles that share no timecode, transcripts and speaker attribution, loudness and audio mastering, cutting, matching two cameras' colour or identifying a log format, and assembling an edit by script in DaVinci Resolve (recommended), Premiere Pro or CapCut (b

Images, Video & Audio

Browse the most popular Claude skills, all images, video & audio skills, or write your own with the skills generator.