One post folder in, one command, and out come a short film, an instructional hosted by AI hosts, a narration, an interactive player and a YouTube-ready folder. What each stage costs on one DGX Spark, and what we still decide by hand.

0:00 Intro
0:31 1. The script
0:50 2. Voices with word timings
1:07 3. The AI hosts
1:28 4. The edit and the render
1:47 5. Quality gates
2:18 6. The YouTube folder
2:36 Still not automatic
2:53 Wrap-up

AI hosts and AI voices: Daisy Lane and Ama Boateng are AI-generated characters, not real people. On-camera segments are AI video lip-synced to ElevenLabs AI voices; the edit, captions and diagrams are rendered with Remotion.

Lip-sync is loose at pauses: across the four host takes the mouth closed at only 6 of 21 pauses between words. On speech it moves in most frames.

Full write-up, settings and measured numbers on the LEJ Video Lab: https://limitededitionjonathan.com/video-lab/from-post-to-youtube-the-lej-video-pipeline
