JD
AI-native creator studio that turns long-form videos into ready-to-publish bite sized 9:16 vertical YouTube Shorts (or 1:1 square Shorts).
Core trick
An LLM picks candidate moments from a long video, then trims filler and reframes each clip to 9:16.
Description
AI-native creator studio that turns long-form videos into ready-to-publish bite sized 9:16 vertical YouTube Shorts (or 1:1 square Shorts). Uses 7 agents in total (including 1 managed) to do this.
README excerpt
# YouTubeShortsAgent
**YouTubeShortsAgent** is an AI-native creator studio that turns long-form YouTube videos and podcast recordings into ready-to-publish **9:16 vertical YouTube Shorts** (or **1:1 square Shorts**).
Designed with a warm, tactile **Hallmark "Hum" aesthetic** for everyday content creators, the app hides complex video-editing timelines behind a 4-step creator flow (**Drop a Video → Pick a Moment → Snip the Fluff → Make it Shine**) and a conversational **Chat with Your YouTubeShortsAgent** assistant that can drive the entire workflow via natural language.
---
## Repository Structure
```text
/
├── server.ts # Express + Vite full-stack server; mounts /api/trends/* & /api/agents/* routes and warms the Cloud Export sandbox on boot
├── firebase-applet-config.json # Google / Firebase OAuth client configuration (used for YouTube Data API v3 direct upload)
├── metadata.json # AI Studio applet metadata & capabilities
├── package.json # Scripts (dev, build, start) & dependencies (@google/genai, @remotion/player, express, lucide-react)
└── src/
├── App.tsx # Root state machine: manages 4-pass pipeline, Test/Reset demo state, sessionStorage persistence, and chat action dispatcher
├── main.tsx # React 19 entry point
├── tokens.css / base.css / index.css # Hallmark "Hum" warm design tokens (Warm Alabaster #FAF7F2, Oa