Guizang Open-Sources a Skill That Turns Codebases Into Product Launch Videos
op7418 · x · 2026-09-28
Guizang (op7418) open-sourced a "product video skill" that packages the workflow he refined while making the CodePilot launch video: hand your codebase to an AI coding tool, tell it what changed, and it produces a software update promo video.
What it does
- Reads recent commits and release notes, picks the changes worth telling, and turns them into audience-friendly selling points
- Prefers wiring in the product's real components, styles and demo states, then drives animation with code, reusing the product's colors, fonts, radii and interaction language
- Rewrites technical changelog language into plain sentences
- Writes original music in code for the piece, adds separate sound effects for interactions, and ducks the music under key cues
- Exports the video plus an editable project
How to use it
- Install: npx skills add https://github.com/op7418/guizang-product-video-skill --skill guizang-product-video-skill
- Then, inside your product repo, tell a skill-capable AI coding tool to make a 45–60s horizontal Chinese promo of the last three weeks of updates, first inspecting the repo's own components and design language and showing key frames before animating and adding sound
- First run checks Node.js, Python, FFmpeg and rendering dependencies
The repo has 475 stars and 44 forks and supports Claude Code and Codex.
More from coding & agent
- A PR arrived saying "I wouldn't expect humans to understand this code" — smlpth · 2026-09-28
- shadcn updates chatbot-template with latest Next.js, AI SDK and fixes — shadcn · 2026-09-28
- Solo dev of 5 years with no reviews: AI agent is now my only code reviewer — Specialist_Agent3599 · 2026-09-28
- New study: coding agents speed up tasks but reduce users' understanding of their own code — boydgraber · 2026-09-28
- Blogger builds an AI trend-radar info assistant on a meta-agent platform — yangyi · 2026-09-28
- Incident agent with persistent memory still scored 0/5 on root-cause in held-out eval — GeethaSriVaraLakshmi · 2026-09-28