erduo1998-cell/agent-motion

erduo1998-cell のプロフィール

記録された最新のソース活動: 2026年10月10日 13:44

SkillsMP カタログ更新: 2026年10月17日 00:09

収集済み skills: 2

GitHub スター: 88
GitHub フォーク: 16

リポジトリの README

README の元ファイル：README.md

README の元ファイルを読む

README 全文を表示

# Agent Motion **Turn talking-head footage into a designed film with a coding agent.** Recording + complete SRT + brief → MP4 + editable Three.js project. [Demos](#demos) · [Get started](#start) · [How it works](#workflow) · [Compatibility](#compatibility) · [Membership & help](#membership-help) **English** · [简体中文](README.zh-CN.md) · [日本語](README.ja.md) · [Español](README.es.md) · [Français](README.fr.md) <img src="docs/images/agent-motion-cover.png" alt="Agent Motion — Words into motion" width="620"> <a id="demos"></a> ## Demos **Greenery, explained** · 60 seconds · Original left / result right <img src="docs/media/greening.gif" alt="Greenery: original on the left, designed film on the right" width="620"> <a href="docs/media/greening.mp4">Watch the clearer MP4</a> <details> <summary>Hybrid training · Expand the 60-second comparison</summary> <img src="docs/media/hybrid-opening.gif" alt="Hybrid training: original on the left, designed film on the right" width="620"> <a href="docs/media/hybrid-opening.mp4">Watch the clearer MP4</a> </details> <details> <summary>Teacher communication · Expand the 60-second comparison</summary> <img src="docs/media/communication.gif" alt="Teacher communication: original on the left, designed film on the right" width="620"> <a href="docs/media/communication.mp4">Watch the clearer MP4</a> </details> Three owner-approved films with matching source timecodes, tracked soft-edged face blur and silent previews. [Versions, excerpts and viewing scope](docs/demo-evidence.md) <a id="start"></a> ## Get started **1. Download and install** You only need **Node.js 22+ and a coding agent**. Download this repository with **Code → Download ZIP**, then extract it; Git is not required. Open the extracted folder and double-click **`start.command` on macOS** or **`start.bat` on Windows**. On Linux, or from your agent's terminal, run: ```sh node scripts/bootstrap.mjs ``` `npm run onboard` runs the same installer. It prepares the project dependencies, local Python environment, Chromium, FFmpeg / ffprobe and the person-matting model, then checks the environment and runs a short matting test. The first run needs an internet connection; let it finish. You do not need to install Python separately or tune a model. [Step-by-step guide and troubleshooting](docs/getting-started.md#english) **2. Give your agent the brief** Put your recording and complete SRT in `inputs/`, open the extracted folder in your coding agent, and send: > Read AGENTS.md, follow the installation guide if needed, then read the two project-local Skills. Create a complete talking-head film using inputs/source.mp4 and inputs/source.srt. Prepare and verify a synchronized person cutout before composition; keep the original speech order and timing. Work in work/my-first-film/. Complete every production stage and inspect the actual picture and audio. Deliver the MP4 and editable project. Respond in English. **Person matting is required for talking-head films.** The installer supplies a local CPU-capable method; a suitable existing person layer or another compatible tool can replace that method. A dedicated GPU is not required. Image generation remains optional. **3. View the result** Ask your agent to open the finished MP4 and editable preview. It can run `npm run serve` and open the film page it created at [work/my-first-film/](http://127.0.0.1:8793/work/my-first-film/). The server root has no preset film. <details> <summary>Checks your agent can run</summary> ```sh npm run doctor -- --matting npm test npm run smoke ``` Doctor checks the matting environment; the rendering smoke performs a real Three.js → H.264 render. Neither replaces reviewing the person edges and the finished film. Installation and these local checks do not call paid generation services. </details> <a id="membership-help"></a> ## Membership & help Want project resources, video examples, or written Q&A with the author? Erduo’s SURGE (无限涌动) offers Seed Membership with lifetime membership status, project resources, video examples, ongoing updates, and community text Q&A. You can also connect your own AI to the member library within your access rights. Available resources, benefits, and prices are listed on the website. [Explore membership](https://erduo.art/site/join) · [Sign up / sign in and purchase](https://login.erduo.art/purchase/checkout) · [Discuss custom services](https://erduo.art/site/business) Member text Q&A excludes remote installation, hands-on engineering troubleshooting, and one-to-one production work. Video production, deployment, training, and special licensing require a separate scope and quote. Membership does not replace this repository’s license or automatically grant commercial rights. Contact the author through the website. Existing members can open the [member library](https://login.erduo.art/studio/#library). <a id="workflow"></a> ## How it works Analyze the complete subtitles, then prepare assets and the person layer, compose type and space, animate continuously, mix sound, and review the output. <img src="docs/images/production-flow.png" alt="Recording, complete SRT and brief pass through analysis and five production stages to an MP4 and editable project" width="940"> One continuous author develops the opening and hardest passage before expanding the film. The agent reads each stage, preserves speech timing and inspects the actual picture and sound. <details> <summary>Design principles: meaning, people, typography and continuous motion</summary> <img src="docs/images/design-system.png" alt="Meaning first, people in frame, type with space, and motion that connects" width="940"> - **Meaning first**: choose actions that explain the spoken idea; keep evidence and timing intact. - **A person in space**: keep a synchronized person layer, visible depth and intentional framing. Outline only when replacing the original background. - **Typography with a role**: compare fonts in real compositions; establish hierarchy, readable landings and reading time. - **Continuous movement**: carry objects and attention between ideas. Validate a continuous sample before expanding the film. </details> <a id="compatibility"></a> ## Compatibility **Agents**: Codex, Claude Code, Gemini CLI, Cursor and GitHub Copilot have entry instructions for the same local Skills. Other agents can read `AGENTS.md` directly. No Codex-specific API is required. Hosts need file, terminal, browser and audiovisual inspection capabilities. **Platforms**: all six Windows, macOS and Ubuntu × Node 22/24 CI jobs passed automatic setup, real CPU matting and video-rendering checks on September 20, 2026. Full-film production is evidenced in Codex; the other clients have not each completed an independent full-film test. [Compatibility and test evidence](docs/compatibility.md) ## Explore further - [Architecture](docs/architecture.md) — the two Skills, toolchain and repository structure. - [Reference analysis library](reference-library/analysis/README.md) — 43 cases, 109 mechanism windows and five tutorial methods; text / JSON only, without reference footage, audio, frames or complete transcripts. - [Contributing](CONTRIBUTING.md) — how to make and verify changes. READMEs are available in five languages; canonical Skills are maintained in Chinese. Agents can work in your language. Check fonts, glyphs, line breaks and reading time for each film; automatic dubbing is not included. ## License **Noncommercial use. Commercial use requires prior written permission.** Project-owned code, Skills, documents and analysis use **Motion Craft Community License 1.0**, a custom license based on Apache-2.0 terms with noncommercial restrictions. Fonts and dependencies retain their licenses; demo footage is excluded from the project grant. [Full license](LICENSE) · [Commercial permission](COMMERCIAL-LICENSE.md) · [Third-party notices](THIRD_PARTY_NOTICES.md)

リポジトリの README を読み込んでいます…

このリポジトリの skills

分類待ち

skill

職業分類

説明

更新

motion-craft

未分類

制作或改善完整动效影片，将内容、排版、空间、交互与镜头的设计判断落实到实际输出；适用于中英混排、MG、品牌产品片与连续UI，沿用AE、Blender、Three.js等项目宿主。

原文の言語: 中国語

2026-10-10

srt-reference-to-three

未分類

根据完整 SRT、原片与参考形成可直接交给 Motion Craft 的内容分析产物；保留语义、原时间、重点、证据及用户修订，并衔接本项目完整视频和口播包装制作。

原文の言語: 中国語

2026-10-10

収集済み skill 2 件中 2 件を表示しています。

AI アシスタントでインストール

このプロンプトをコピーして、使用中の AI アシスタントに送ってください。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/erduo1998-cell/agent-motion.