showreel

showreel

A Claude Code skill that produces Korean motion-graphics videos from a single request — concept, research, script, narration, music, images, render, critique and rework, mix and QA.

English · 한국어

Highlights from a 2-minute promo for showreel, made with this skill. The full video below has sound.

---

Demo

The full 2-minute video was produced by this skill from one prompt: four ElevenLabs voices, generated music, sound effects, AI images, 22 motion techniques and an automated QA pass.

▶ Download the full demo (1080p, 2:00, 31 MB)

한국어 소개

showreel은 Claude Code에서 한국어 모션그래픽 영상을 만드는 스킬입니다. 정해 둔 무드나 템플릿을 고르지 않고, 요청마다 콘셉트 후보를 여러 개 세워 가장 강한 것을 고른 뒤 구간마다 아트 디렉션을 새로 정해 만듭니다. 사용자에게는 사용자만 아는 사실(대상, 꼭 넣을 정보, 자료 위치)만 묻고, 무드·길이·나레이션 여부 같은 창작 결정은 Claude가 내립니다. 조사, 원고, ElevenLabs 나레이션, 음악과 효과음, 이미지 생성, HTML·WebGL 모션 엔진 렌더, 비평과 재작업, 믹스, 자동 검수를 모두 거칩니다.

화면 전환은 나레이션 단어와 음악 박자에 프레임 단위로 맞춥니다. 완성본은 받아쓰기(STT)로 발음을 다시 확인하고, 음량·섬광·멈춘 화면을 검사한 뒤 납품합니다. 인물 프로필, 기업 홍보, 행사, 교육, 티저, 음악 기반 영상 등 실제 제작 13건에서 쓴 방법을 스크립트와 문서로 정리했습니다.

자세한 설명은 한국어 README에 있습니다.

Table of contents

• Demo
• Overview
• Features
• How it works
• Requirements
• Installation
• Quick start
• Defaults
• Script reference
• Repository layout
• Configuration
• Troubleshooting
• Limitations
• Contributing
• Security
• License
• Acknowledgments

Overview

showreel is an Agent Skill for Claude Code. It turns a short request such as "Make a 60-second promo for our product launch, cinematic, with narration" into a finished, delivery-ready video:

• a master MP4 and a lighter MP4 for sharing,
• a poster JPG and an SRT subtitle file,
• a production note listing sources, settings, open checks and the commands to regenerate every asset.

The skill is built around one request: "Show how amazing a motion graphics designer you are — a dynamic one-minute Korean motion graphics video, like a showreel for a résumé. Give it everything." The video itself has to prove the craft. There is no default mood or template: Claude writes several concept candidates, scores them, gives each segment its own art direction (medium, palette, type, motion character), builds the picture in code (DOM, Canvas 2D, WebGL shaders, three.js), and rebuilds the weakest segment after every draft until a written critique keeps every segment.

The skill is written for Korean content (Korean typography, Korean narration and pronunciation rules), but the engine, renderer and audio tools are language-neutral.

Features

How it works

flowchart LR
    A[0 Preflight] --> B[1 Read request<br/>ask only facts]
    B --> K[2 Concept<br/>3+ candidates scored]
    K --> C[3 Research & assets]
    C --> D[4 Storyboard & script]
    D --> E[5 Audio first<br/>music · SFX · narration → timeline]
    E --> F[6 Images]
    F --> G[7 Style frames<br/>per segment]
    G --> L[8 Scenes]
    L --> H[9 Draft & critique loop<br/>rebuild weakest segment]
    H --> I[10 Final render · mix · finish]
    I --> J[11 Automated QA & delivery report]

Audio is produced before the picture. timeline.py converts narration word timings and music hits into named cues, and scenes read those cues with REEL.cue('name') instead of hard-coded seconds. When a narration line is regenerated, the picture follows without edits.

Long-running steps (research, TTS, music, image generation, final render) run in the background while Claude continues with the next step.

Requirements