andyhuo520/vibe-knowledge
andyhuo520/vibe-knowledge has 672 GitHub stars and 108 forks. GitHot measured +13 stars in 24 hours. Last checked 2026-11-15 23:43 UTC. Last push 2026-10-28.
Measured star growth
- in 24h
- +13
- in 7 days
- Update delayed
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
Vibe 知识大赏:一个知识点 → 一条有旁白、每句有出处的精品知识短片(Claude Code skill · 44 套风格 · 代码逐帧成片)
Developer ToolsREADME.en.md
Originalvibe-knowledge
One idea → one narrated, fully sourced, cinematic explainer short.
A Claude Code skill that takes a topic through research → script → voice → code-rendered video (1080p60) → QA → publishing kit, all inside the chat.
中文 · Samples · Styles · Audio · Renoise · Install

Excerpt of the style flash reel. Full 42 s: style-flash-42s.mp4
Samples
Click a poster to play (videos live in Release v1.0.0). Narration and captions are in Chinese; labels under each poster give the template · style · tier. Every frame is orchestrated by Claude Opus and rendered frame-by-frame in code; "enhanced" shots use Renoise for AI images and motion clips, always labelled as AI illustrations on screen.
Style reels (each style performs the same opening, for side-by-side comparison):
- 最早 7 套(古风、星空光点、暗金线描、水墨纸本、像素、字幕新闻、视觉调查)
- 剪辑与纪实
- 补空白家族 8 套
- 增强档:敦煌、青花、木刻、半调拼贴
- 第五批 8 套
- 第六批 8 套
- 增强档:长卷、铜版博物
- 田园像素
- 第九批 5 套
- 手绘过程(手 + 笔画给你看)
44 styles · 8 families

| Family | Count | Styles (each STYLE.md covers palette, type, camera language, best-fit topics) |
|---|---|---|
| Eastern | 7 | Gufeng, Ink & Paper, Dunhuang Mural, Blue & White, Woodcut, Minimal Zen, Handscroll |
| Dark concept | 7 | Starfield, Dark Gold, Dark Math, Neon Cyber, Keynote, Aurora Glass, Art Deco |
| Game | 4 | Pixel, Rpg Dialogue, Card Battle, Cozy Farm |
| Editorial & documentary | 5 | Caption News, Visual Investigation, Retro Desktop, Light Table, Tv News |
| Bright pop | 5 | Variety Captions, Memphis Pop, Bullet Journal, Vaporwave, Clay Kawaii |
| Diagram & design | 8 | Chalkboard, Blueprint, Isometric, Swiss Kinetic, Metro Map, Isotype, Copperplate, Hand-Drawn Process |
| Print & paper | 5 | Risograph, Halftone Collage, Ransom Note, Editorial, Park Poster |
| Cinematic | 3 | Title Sequence, Silent Film, Constructivist |
Default voice & BGM
- default-voice-demo.m4a: the default narrator, an AI voice made with Xiaomi MiMo voice design (not a real person).
- mix-demo-voice-plus-bgm.m4a: final mix (voice + BGM + SFX).
- vibe-bgm.m4a: the default BGM. It is mandatory: the mixer auto-fits it to the film length, disables other full-length scores, and calibrates it to sit 10 dB under the voice while speaking. Swapping it requires an explicit user request (
bgm: true, user_request: "...").
How an episode is made
Pick a topic → choose one of 6 narrative templates, a length (four tiers from 45 s to 5 min, each stating how deep it can go and what gets cut), a style and a tier → sourced fact sheet → script + adversarial review → MiMo voice with word-level alignment → film.js with the know.js component kit → frame-by-frame render → mix (mandatory BGM) → cover as first frame → QA (contact sheet, OCR read-back vs facts, loudness, freeze detection, BGM detection) → publishing kit. Three explicit user checkpoints; any paid generation is quoted first.
Two tiers share one film.js: base (0 credits: hand-written SVG layers, particles, data components) and enhanced (only the shots code can't do well go to Renoise).
Renoise: the generation platform for the enhanced tier
Renoise is an AI video, image and audio generation platform that plugs straight into Claude Code, Codex and other agents over MCP. One account gives you Midjourney, GPT Image, Nano Banana, Seedance 2.5, Wan 3.0, Kling, Grok Video, Lyria, ElevenLabs SFX and more, right inside the chat.
vibe-knowledge only uses it for the enhanced tier, for the few shots code can't do well: realistic period scenes, the one to three motion shots viewers should remember, and the occasional sound effect. Everything else is still drawn in code. The base tier needs no Renoise at all and costs 0 credits.
- Quote first, generate only on your OK: every episode lists shots, models, prompts and the
estimate_costtotal, and nothing is submitted until you approve in the chat. - Predictable cost: episode 01 (enhanced) used 246 credits (10 Midjourney images, 2 Wan motion shots, music, 5 SFX). Enhanced levels are roughly light ≈25–40, standard ≈250–300, flagship ≈600–700 credits.
- Renoise-generated imagery is labelled as an AI illustration on screen.
No account yet? Sign up at renoise.ai. Sign-in happens in the browser; no API key to copy. Then connect your agent:
| Agent | How to connect |
|---|---|
| Claude Code | /plugin marketplace add ArcoCodes/renoise-plugins-official, then /plugin install renoise@renoise-plugins-official (browser sign-in on first use) |
| Claude desktop / claude.ai | Settings → Connectors → Add custom connector → name Renoise, URL https://renoise.ai/api/public/mcp |
| Codex | codex mcp add renoise --url https://renoise.ai/api/public/mcp, then codex mcp login renoise |
| OpenClaw | openclaw plugins install renoise --marketplace ArcoCodes/renoise-plugins-official |
| Cursor, Windsurf, Cline and other remote-MCP clients | Add a remote MCP server at https://renoise.ai/api/public/mcp (typically {"mcpServers":{"renoise":{"url":"https://renoise.ai/api/public/mcp"}}}) |
Latest instructions: renoise.ai/plugin. Once connected, ask your agent "check my Renoise balance" to confirm.
Install
git clone https://github.com/andyhuo520/vibe-knowledge.git
cp -R vibe-knowledge/skills/vibe-knowledge vibe-knowledge/skills/ai-teardown ~/.claude/skills/
Requirements: a Python with numpy scipy pillow playwright (set CINE_PY), ffmpeg, macOS Vision + swiftc for OCR QA, a Xiaomi MiMo key in MIMO_API_KEY or ~/.config/ai-teardown/mimo.env, and optionally the Renoise MCP for the enhanced tier (see Renoise above). Then ask Claude Code: "Make a Vibe knowledge episode: why does the digit 1 give fake books away?"
Acknowledgements
We studied and learned from open-source projects including lemo-opuscar, huashu-art-motion, paper2video, Paper-Cut, LuZhong-Li/vibe-knowledge-video-skill and earshot. What we took from each one, where it is used and under which licence are listed in ACKNOWLEDGEMENTS.md. Corrections are welcome via issues.
License
Code: MIT. Sample videos, AI imagery, default BGM and voice: see NOTICE.md. Third-party components: THIRD_PARTY.md.
By @Berryxia · orchestrated with Claude Code (Claude Opus)







