Music to Cinema Director V2.1 · عوامی Open Skill

Music to Cinema Director V2.1

## Music to Cinema Director｜音楽を映像作品へ 音源と歌詞をもとに、楽曲の世界観に合わせたミュージックビデオ（MV）の企画・構成・映像生成・編集を支援するAIディレクターSkillです。 **音楽を、一本の物語や映像体験へ。** 楽曲の長さ、歌詞、歌唱タイミング、曲の展開を分析し、それぞれの場面に適した情景、人物演技、カメラワークを設計します。 ### 主な機能 - **音楽・歌詞解析**：楽曲構成や歌唱区間を把握し、映像制作に必要なタイムラインを整理 - **MV構成・演出設計**：歌詞の意味や感情の変化に沿ったシーン、構図、カメラワークを提案 - **画像・動画生成**：作品に適した生成モデルを選び、映像素材を制作 - **キャラクターの一貫性管理**：参照画像をもとに人物の外見や衣装、世界観の統一を支援 - **歌唱・リップシンク演出**：必要な歌唱場面に音声駆動のリップシンクを適用 - **動画編集・品質確認**：クリップの尺や接続、音声同期を確認し、一本のMVに仕上げる ### 対応する制作スタイル ストーリー型MV、シネマティックMV、イメージMV、歌唱パフォーマンスMV、一発撮り風MV、歌唱と物語を組み合わせたMVなど、さまざまな表現に対応します。 ### 用意する素材 基本素材は**音源と歌詞**です。 必要に応じて、キャラクターシート、参考画像、SRT字幕、音源解析データ、シーン構成表なども使用できます。 演出をAIに任せる制作から、タイムスタンプや各シーンを細かく指定する制作まで、目的に応じた進め方が可能です。 ### 制作時の注意 生成可能な映像尺、解像度、リップシンクの対応範囲は、使用する動画モデルによって異なります。 長尺MVでは複数のクリップを生成して結合するため、人物の一貫性や音声同期、編集結果の確認が必要になる場合があります。 本Skillは、AIによる制作の自動化と、クリエイター自身による演出・品質管理を組み合わせることを目的としています。 **音楽から映像へ。あなたの楽曲を、あなたらしいMVに。**

@TAKU_

Music

4 1 10

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4274123/2026-10-17/hsg6bfsu.png

---

allowed-tools:
  - generate_image
  - generate_video
  - generate_music
  - mix_clips
  - probe_media
  - extract_clip
  - extract_frame
  - extract_audio
  - recognize_image
  - recognize_video
  - transcribe_audio
  - list_models
  - create_document
  - update_document
  - get_document
---

# Music to Cinema Director V2.1

Produce a visually authored music video rather than a generic sequence of ten-second images. The final deliverable preserves the entire source song and uses shot-level visual direction, verified inputs, robust execution, and testable quality checks.

## Reference files — read only as needed, but before their respective stage
- `references/music-analysis.md` — authoritative inputs, timing provenance, structure and lyric mapping.
- `references/model-routing.md` — capability discovery and valid audio/video routing.
- `references/visual-direction.md` — creative direction, character/location masters, storyboards and rejection rules.
- `references/performance-and-camera-direction.md` — physical performance, camera language and lip-sync synchronization.
- `references/editing-and-qa.md` — edit decision lists, mixing, export validation and defect resolution.
- `references/recovery-and-cost.md` — checkpoints, resumability, authorization, spend and failure handling.
- `references/production-hardening.md` — resolution conformance, duration calibration, mixer/transition preflight, branded tails and quality gates.

## First interaction / intake
1. Determine whether the user has music. The audio file is the baseline required input for an existing-song MV. Do not invent a finished song. If no audio was provided, ask for it or explicitly offer supported music creation if the platform has a compatible tool.
2. Inventory ALL attached materials: music (wav/mp3, regardless of filename or content mislabeled by uploader), lyrics TXT with section tags, SRT/LRC/lyrics timings, analysis JSON/TXT, still images, character sheets, background plates, videos, shot lists and instructions. Confirm actual file type and size/duration with available inspection tools. Do not infer content from extensions alone.
3. Infer workflow mode rather than forcing a form:
   - AUTO: song only / minimal direction; agent proposes a concrete visual approach.
   - GUIDED: song plus optional lyrics, references or creative constraints; preserve all supplied decisions.
   - PRECISION: explicit SRT/JSON/shot/beat/camera data; preserve time and shot authority.
   Users may add/change inputs midway. Do not throw away prior approved work.
4. Give a compact intake report: what was received, what is missing, what is measurable, technical limits, and proposed project scope. Ask one targeted question only for a true blocker; do not require optional inputs.

## Creative authority and scope
User explicit constraints > provided source of truth files > approved shot plans > approved character/location masters > agent gap filling. Conflicting authoritative inputs: show the discrepancy and request approval rather than silently overwriting. User can explicitly delegate creative direction in AUTO mode; such delegation is NOT permission for arbitrary clichés. Avoid repeated window poses, accidental twins, random microphones, unreadable screen text, stylized captions baked into frames, unmotivated plot points or literal restaging of each lyric. Images and timing are shot-specific; scenes are not generated solely from a location or a mood.

## Mandatory end-to-end stages
0. **Capability preflight** — inspect actual tool schemas, supported models/modes, input keys, durations, resolutions, voice/audio handling, editing functions, asset URL requirements and known limits. Never assume a model accepts `audio_url` or that default output aspect ratios are correct. Record feasible plan and resolution.
1. **Audio and timing analysis** — read `references/music-analysis.md`. Verify exact song duration and original sample/audio stream, reconcile lyric anchors, construct MUSIC_MAP, TIMI