virtual-singer-mv-script

🎬 Virtual Singer MV Script Generator

Generate a complete MV storyboard script for a virtual singer — from lyrics/audio to an executable video production blueprint

Trigger Words

Generate MV script
MV storyboard script
Virtual singer video script
mv script
Storyboard design
Analyze audio to generate MV

Input (Choose any one method)

Method	Parameter	Required	Description
Audio file	`audio`	✅	MP3/WAV/M4A file path (Recommended, precise sync)
Lyrics only	`lyrics`	✅	Lyrics text
Character card	`character_card`	❌	Singer character card path (default: Chloe)
Style	`style`	❌	cinematic / ethereal / anime / realistic
Number of scenes	`scenes`	❌	Default 9

Recommended method：python3 scripts/audio-analyzer.py <audio_file> -s 9 -v → Automatically analyze duration, BPM, energy curve, and generate a precise timeline

Output

Complete storyboard in JSON format — can be directly integrated into the video generation pipeline
Readable Markdown script — for human review
Emotion curve chart — emotion_score 0-10 trend
Timeline alignment — precise to the millisecond, synchronized with audio

Workflow

音频文件 / 歌词文本
        │
        ▼
┌─────────────────────────────────┐
│  0. 音频分析 (audio-analyzer)   │  ← 新增第一步
│     - mutagen 读取真实时长      │
│     - librosa BPM 检测          │
│     - 能量曲线 → 场景时间节点    │
│     - 输出: scene_timeline.json  │
└──────────────┬──────────────────┘
               │ (无音频时跳过，用歌词估算)
               ▼
┌─────────────────────────────┐
│  1. 歌词解析 (lyrics-parser) │
│     - 结构识别 (Verse/Chorus/Bridge) │
│     - 情绪曲线计算 (0-10)    │
│     - 时间轴对齐（精确到秒）  │
└──────────┬──────────────────┘
           │
           ▼
┌─────────────────────────────┐
│  2. 场景分配 (scene-assigner) │
│     - 情绪→场景类型映射      │
│     - Chloe 专属场景库       │
│     - 运镜设计              │
│     - 首尾帧衔接设计         │
│     - 音频同步校正           │
└──────────┬──────────────────┘
           │
           ▼
┌─────────────────────────────┐
│  3. 模型选择 (model-selector) │
│     - 场景复杂度评估         │
│     - 视频生成模型推荐       │
│     - 参数配置              │
└──────────┬──────────────────┘
           │
           ▼
  完整 MV 分镜剧本 (JSON + MD)

Quick Start

Step 0: Audio analysis (recommended)

python3 ~/clawd/skills/virtual-singer-mv-script/scripts/audio-analyzer.py \
  /path/to/song.mp3 \
  -s 9 -v -o /tmp/song_analysis.json

Output: duration, BPM, precise start and end times of 9 scenes

Step 1: Generate the complete script

cd ~/clawd/skills/virtual-singer-mv-script

# 有音频 → 精准模式
python3 scripts/audio-analyzer.py "/path/to/song.mp3" -s 9
python3 scripts/lyrics-parser.py --lyrics "..." --bpm 88
python3 scripts/scene-assigner.py --parsed /tmp/lyrics_parsed.json \
  --timeline /tmp/song_analysis.json
python3 scripts/merge-output.py \
  --lyrics /tmp/lyrics_parsed.json \
  --scenes /tmp/scenes_assigned.json \
  --output mv_script.json

Scene time allocation rules (audio sync mode)

Section	Duration proportion	Emotion strategy
Intro (0-10%)	10%	Calm 3.0
Verse 1 (10-25%)	15%	Gradual rise 4.5
Transition (25-40%)	15%	Build-up 6.0
Chorus 1 (40-55%)	15%	Outburst 8.0
Chorus 2 (55-70%)	15%	Sustain 8.5
Climax (70-82%)	12%	Peak 9.0
Bridge (82-92%)	10%	Continuation 8.5
Finale (92-98%)	6%	Highest 10.0
Outro (98-100%)	2%	Dissipation 2.0

Key principle: All scene durations are calculated based on the actual audio duration and must not be estimated using fixed values

Frame file specifications

Frame naming:scene{NN}_{first|last}.png

scene01_first.png — Scene 1 first frame
scene01_last.png — Scene 1 last frame

Required elements (each frame prompt must include):

A beautiful young East Asian woman, long black wavy hair,
pink cherry blossom hair clip, white lace dress,
holding cherry blossom branch near her lips singing, fair skin

Video model selection rules

Scene type	Recommended model	Reason
Static character + natural light	kling-1.6	High cost-effectiveness, stable static shots
Dynamic camera movement + character	kling-2	Strong understanding of camera movement
360° rotation/complex effects	veo-3.1	Best for particle/rotation effects
Long take + narrative	sora-2	Long-duration consistency
Guofeng/dance	seedance-3.0	Smooth Eastern aesthetics

Notes

Audio priority — When there is audio, audio-analyzer.py must be used to analyze the actual duration
First and last frame continuity — The visual elements of the last frame must appear in the first frame of the next scene
Emotional curve coherence — Sudden jumps are not allowed (unless required by the music design)
Exact total duration — The sum of all storyboard durations = the total audio duration
BPM sync — Transition beats must land on the beat

About this skill