generate-video

Generate Video Skill

This is a group of skills responsible for the automatic generation of product explanation videos.

Activation Contract

This skill is user-invocable: false and disable-model-invocation: true . It is not assumed that Claude will automatically select it based on a normal user utterance (e.g., "make a video").

• Explicit /generate-video command, or use it only when invoked internally from a parent media workflow.
• allowed-tools is for Claude Code. In Claude, you can use AskUserQuestion for user confirmation.
• In the Codex mirror, the equivalent confirmation is reinterpreted as input in a plan mode corresponding to request_user_input . If that tool is not available during normal execution, present the default values and proceed without stopping, and confirm only important undetermined matters with the user.
• disable-model-invocation: true means "the model does not start on its own just by looking at the description." If explicitly activated from a slash command or a parent workflow, follow the steps in the body.

---

Overview

/generate-video This skill is used internally by the command. It executes the flow of codebase analysis → scenario proposal → parallel generation.

Feature Details

Feature	Details
Best practices	See references/best-practices.md
Codebase analysis	See references/analyzer.md
Scenario planning	See references/planner.md
Parallel scene generation	See references/generator.md
Visual effects library	See references/visual-effects.md
AI image generation	See references/image-generator.md
Image quality assessment	See references/image-quality-check.md

Prerequisites

• Remotion is already set up (/remotion-setup）
• Node.js 18+
• (Optional)GOOGLE_AI_API_KEY - for AI image generation

/generate-video Flow

/generate-video
    │
    ├─[Step 1] 分析（analyzer.md）
    │   ├─ フレームワーク検出
    │   ├─ 主要機能検出
    │   ├─ UIコンポーネント検出
    │   └─ プロジェクト資産解析（Plans.md, CHANGELOG等）
    │
    ├─[Step 2] シナリオ提案（planner.md）
    │   ├─ 動画タイプ自動判定
    │   ├─ シーン構成提案
    │   └─ ユーザー確認
    │
    ├─[Step 2.5] 素材生成（image-generator.md）← NEW
    │   ├─ 素材必要判定（イントロ、CTA等）
    │   ├─ Nano Banana Pro で2枚生成
    │   ├─ Claude が品質判定（image-quality-check.md）
    │   └─ OK → 採用 / NG → 再生成（最大3回）
    │
    └─[Step 3] 並列生成（generator.md）
        ├─ シーン並列生成（Task tool）
        ├─ 統合 + トランジション
        └─ 最終レンダリング

Execution Steps

1. User /generate-video Run
2. Remotion setup check
3. analyzer.md Analyze the codebase with
4. planner.md Propose a scenario + user confirmation with
5. generator.md Parallel generation with
6. Completion report

Video types (by funnel)

Type	Funnel	Target length	Auto-detection condition	Structure core
LP/ad teaser	Awareness→Interest	30-90 seconds	New project	Pain→Result→CTA
Intro demo	Interest→Consideration	2-3 minutes	UI change detected	Complete one use case
Release notes	Consideration→Conviction	1-3 minutes	CHANGELOG updated	Emphasis on Before/After
Architecture explanation	Conviction→Decision	5-30 minutes	Large-scale structural change	Real operation + evidence
Onboarding	Continuation/Utilization	30 seconds-several minutes	Initial setup	Shortest path to the Aha experience

Details: references/best-practices.md

Scene templates

90-second teaser (for LP/advertising)

Time	Scene	Content
0-5 seconds	Hook	Pain or desired result
5-15 seconds	Problem+Promise	Target users and promise
15-55 seconds	Workflow	Signature workflow
55-70 seconds	Differentiator	Basis for differentiation
70-90 seconds	CTA	Next Move

3-Minute Intro Demo (for review)

Time	Scene	Content
0-10 seconds	Hook	Conclusion + pain point
10-30 seconds	UseCase	Use case declaration
30-140 seconds	Demo	Complete run-through on the actual screen
140-170 seconds	Objection	Address one common concern
170-180 seconds	CTA	Call to action

Common scenes

Scene	Recommended time	Content
Intro	3-5 seconds	Logo + tagline
Feature demo	10-30 seconds	Playwright capture
Architecture diagram	10-20 seconds	Mermaid → animation
CTA	3-5 seconds	URL + contact information

Detailed template: ${CLAUDE_SKILL_DIR}/references/best-practices.md

Audio synchronization rules (important)

For videos with narration, strictly follow the rules below:

Rules	Value
Audio start	Scene start + 30f (wait 1 second)
Scene length	30f + audio length + 20f margin
Transitions	15f (overlap with adjacent scene)
Scene start calculation	Previous scene start + previous scene length - 15f

Pre-check: ffprobe Check the audio length with before designing the scene

Details: ${CLAUDE_SKILL_DIR}/references/generator.md

BGM support

Item	Recommended value
With narration	bgmVolume: 0.20 - 0.30
Without narration	bgmVolume: 0.50 - 0.80
File placement	public/BGM/

Details: ${CLAUDE_SKILL_DIR}/references/generator.md

Subtitle support

Rules	Value
Subtitle start	Same as audio start
Subtitle duration	Audio length + 10f
Fonts	Base64 embedding recommended

Details: ${CLAUDE_SKILL_DIR}/references/generator.md

Visual effects library

A collection of effects for high-impact videos:

Effects	Use
GlitchText	Hook, title
Particles	Background, CTA convergence
ScanLine	Analyzing presentation
ProgressBar	Parallel processing display
3D Parallax	Card display

Details: references/visual-effects.md

Notes

• If Remotion is not set up, /remotion-setup guide them to
• The number of parallel generations is automatically adjusted according to the number of scenes (max 5)
• Generated videos are out/ output to the directory
• AI-generated images are out/assets/generated/ saved to
• GOOGLE_AI_API_KEY If not set, skip image generation (use existing assets or placeholders)