English | 简体中文
Ideas, brought to light.
Give Codex a question. Knowledge Easy researches it, shapes a story, builds the visual explanation,
and brings the finished film into a local review room.
Install · Finished films · Visual system · Lazy Studio · About
Run this in a terminal with Node.js 22.20+ (including npm/npx) and Git installed:
npx -y skills add feicaiclub/knowledge-easy --agent codex --global --yesThis uses Vercel's Skills CLI to download the skill from GitHub and install it for Codex. --agent codex selects Codex; --global makes it available across projects. The skill and its bundled resources go into ~/.agents/skills/knowledge-easy/.
Alternative: install directly from a Codex chat
If you do not have Node.js or prefer to let Codex handle installation, paste this into a Codex chat:
$skill-installer Install https://github.com/feicaiclub/knowledge-easy from the knowledge-easy directory (path: knowledge-easy), with the skill name knowledge-easy.
This is an instruction for Codex's built-in Skill Installer, not a terminal command. Choose either installation method; you do not need both.
Codex detects newly installed skills automatically. If Knowledge Easy does not appear, restart Codex.
Then tell it what you want to make:
$knowledge-easy Make an English explainer about why we rationalize decisions we have already made. Build it around a concrete story. Use the series narrator and visual system, with footage, music, and graphics that follow the argument. Deliver a finished film and open it in Lazy Studio for review.
You can ask for a complete film, topic ideas, a script, or a revision. Dependencies are checked and installed when needed, outside the skill folder. If setup fails on your machine, Codex diagnoses the cause and works through compatible alternatives, explaining meaningful differences in the result.
The film about Xun Yu follows the tension between the man he helped and the dynasty he hoped to preserve. It moves between dramatic footage, geographic decisions, evidence, and reflection.
 Introduce the person and their position |  Make a strategic choice visible |
 Shift attention across the same map |  Put a proposal beside its implications |
 Separate two centers of power |  Bring in names when the story needs them |
The sunk-cost film starts with an ordinary movie ticket and follows the reasoning into work, relationships, and the time still ahead. Real scenes give the explanation a human setting; graphics make the pattern easier to see.
 Money already spent. Time still ahead. |  Show how a decision reinforces itself |
 Put the next choice back in view |  Return to life outside the diagram |
These stills are taken from completed edits. Underlying film and stock imagery retains its original rights; see third-party notices.
The package includes 24 classic editorial designs plus three extensible code-illustration languages. Original scene code can create new subjects, staging and motion; project-owned references and renderers can go beyond the bundle. The shared system covers typography, cards, diagrams, timelines, evidence layouts, and maps. Content, footage, coordinates, and reveal timing adapt to the story. All twelve base designs can be called directly in films and combined by region or time. Custom designs retain a handbook reference; exported frames and complete scene clips are compared before visual acceptance.
 A thought, then its implication |  Routes and geography |
 Evidence and interpretation |  Two ideas on equal visual footing |
 A detail inside its context |  Two moments, one shared scale |
- Open the visual handbook locally after installation. It combines still references with layout and motion guidance.
- Use the component source, renderer guide, and motion contract to build and extend the system.
- Review the actual motion in each new film. A layout preview shows composition; the source and cue parameters define entrances, holds, and exits.
Use graphics when they add understanding. Let the footage and narration carry the rest.
Pause, press M, write a comment, and press Enter. The version, frame, comment, and screenshot stay together.
 Saved comments keep their original frame and version |  Read the project review without leaving the film |
The comparison above uses English and Chinese demos to show the interface. It is not presented as a before-and-after edit.
Notes are written to the film project's review/feedback.md, with version/frame records and linked screenshots. Return to the original Codex chat and say: “My notes are ready. Read the feedback and make the changes.” The current integration works through local files; it does not automatically attach notes to the chat composer.
The screenshots were captured through actual UI interactions in a separate demonstration workspace. Sample comments are labeled as examples, and the original films and feedback records were preserved.
Knowledge Easy uses the approved Qwen3-TTS D reference and body-v1 finishing. The speaker identity stays consistent while emphasis, emotion, pacing, and phrasing follow the subject.
Listen to the approved English pilot · Mandarin reference · Voice guide
The English pilot was approved through user listening. Transcription checks content and timing; it does not certify accent or naturalness.
| Stage | What happens |
|---|---|
| Research and writing | Develop conversational narration and visual reasoning together; preserve the argument when shortening. |
| Material and shot planning | Source cinematic footage and imagery, generate illustrations, or build editable graphics to explain the subject. |
| Narration | Use the fixed reference; generate complete passages and review names, phrasing, and endpoints. |
| Editing and graphics | Call the shared renderer and adapt the composition to the explanation. |
| Sound | Use separate music cues, purposeful effects, transitions, and silence while keeping speech clear. |
| Review and revision | Inspect the export, read exact-frame notes, preserve approved work, and deliver a new version. |
The quality floor comes from completed films and repeated revisions. The topic still determines the story, duration, rhythm, and creative treatment.
knowledge-easy/
├── SKILL.md One skill: the production workflow
├── references/ Production guides and interfaces
├── scripts/ Setup, voice, transcription, editing, review tools
└── assets/
├── packaging/ Renderer, fonts, maps, static visual references
├── voice/ Fixed reference, profiles, pronunciation
└── lazy-studio/ Review interface
The package contains one skill. Runtime dependencies, model weights, API keys, film projects, and demo videos are not bundled. The public handbook uses still images; production motion remains in the source code.
The default stack preserves the approved workflow: Qwen3-TTS through MLX Audio, MLX Whisper large-v3, HyperFrames, FFmpeg, and the shared renderer. Exact versions and model revisions are in runtime-lock.json.
The original narration and transcription ran on Apple Silicon Macs. Other machines may need a compatible implementation, which Codex investigates when the skill is used. Local model downloads can be large, and synthesis speed depends on the machine. Optional music and effects services use your own account and credits. See the runtime guide.
The original generic component catalogs can be rebuilt with licensed local footage using catalog --media-dir; raw third-party clips are not distributed. See demo media.
Open an issue with your OS, hardware, the failing step, and a redacted error message. For visual changes, show the entrance, reading order, hold, and exit; update the component and handbook together.
Original code and documentation are MIT licensed. Fonts retain their SIL Open Font Licenses, and map geometry comes from Natural Earth. Third-party imagery visible in screenshots is not relicensed under MIT. Third-party notices.
The workflow uses Qwen3-TTS, MLX Audio, MLX Whisper, HyperFrames, GSAP, FFmpeg, and Pedalboard. Lazy Studio builds on FEICAI's existing review tools and UI kit.
Feicai (废才), founder of Feicai Club (废才俱乐部), is an AI builder and educator focused on agents, software, and filmmaking. Each workflow is developed through real projects, then the reusable parts are shared as open source.
- Author of 驱动 AI 提示词：指令入门与场景应用 (Tsinghua University Press), a book on prompting AI and applying it to real work.
- Creator of the Brutally Honest PM (毒舌产品经理) and Seedance Storyboarder (Seedance 分镜师) series.
- More projects: Ad Storyboarder, previz, video-spec-builder, and the DeepSeek Harness whitepaper.
| 🌐 Website | www.feicaiclub.cn |
| 𝕏 | @zhu185178 |
| @feicaiclub | |
| 📺 Bilibili | 废才俱乐部Club |
| 📕 Xiaohongshu (RED) | 废才俱乐部 |
| 🎵 Douyin | search for 废才俱乐部 |
| 💬 WeChat Official Account | search for 废才俱乐部 |
