slides2video — Autonomous Lecture Pipeline

3-minute complete lecture. Zero editing. Zero variance. The web page is already A+, so every capture is A+. 100 videos = 100 identical quality.

Core Paradigm

Web page

= After Effects. HTML/CSS/JS animations replace motion graphics.

Playwright

= Director. Clicks, scrolls, runs JS — exactly like a human.

CDP

= Camera. Browser records itself internally. No Xvfb. No x11grab.

TTS + REAPER

= Voice actor. Park씨's tone, humanized. Megaphone / bathroom / stadium presets.

P0 Script Generation script_generator.py

Claude scans the page + loads repo context → auto-generates narration in Park씨's style

P1 CDP Recording extractor.py

browser.new_context(record_video_dir=) — Chrome records itself. JS animations fully preserved.

P2 TTS Parallel Render tts_humanizer.py ✅

edge-tts → REAPER JSFX 6-stage chain → environment presets → WAV per step

P3 Assembly + Mix renderer.py

webm + WAV timeline merge → BGM mix → 9:16 1080×1920 encode

P4 YouTube Upload upload_youtube.py

Auto title / description / Shorts tags → YouTube Data API v3 → done

Lane A — Human

Park씨 on camera Real face · real voice · real handwriting · APK overlay kit

Brand · Trust · 1~2×/week

Lane B — Auto

Claude as Park씨 Playwright as hands · TTS as voice · CDP as camera

Factory · Algorithm · daily

3min Shorts ← auto pipeline · complete lecture

├── satisfied here → subscribe · repeat

└── want more → 15min handwriting lecture

→ loyal subscriber

One command. Web page becomes a broadcast.