ve5 — Automated Essay Video Generation Pipeline
A pipeline that, given just a single topic keyword, autonomously handles everything from script writing, narration, subtitles, B-roll, visual design, Remotion rendering, and thumbnails to scheduled YouTube uploads.
Duration
In-house development
Team
1 person
01 // OVERVIEW
Project Overview
A pipeline that fully autonomously mass-produces essay videos for YouTube. Given just a single line of theme keywords, it runs the entire process end to end: script generation, narration, subtitle timing, automatic B-roll collection, visual design, Remotion rendering, thumbnail generation, and scheduled YouTube uploads. Output is 1920x1080 Full HD at 30fps, with two length modes spanning 7 to 16 minutes.
02 // KEY FEATURES
Key Features
JSON-Driven Design Centered on timeline.json
All segment definitions, start times, display modes, B-roll assets, and Remotion component props are consolidated into a single timeline.json. Rather than adding a new Python script per video, general-purpose scripts read and process the JSON - a JSON-driven design.
From Automatic B-Roll Collection to Remotion Rendering
Automates the entire process: script generation -> narration -> word-level timestamps -> automatic B-roll downloads -> visual design -> Remotion rendering -> thumbnail generation. Also supports scheduled YouTube uploads running across 3 accounts in parallel.
Role Separation via a Three-Layer Architecture
Separated into three layers: the skill layer (Claude judgment), the Python script layer (mechanical processing), and the Remotion layer (rendering). Rendering logic is consolidated into Essay.tsx / ThumbnailJp.tsx, so the pipeline and UI can be maintained independently.
03 // TECHNICAL DEEP-DIVE
Technical Highlights
Solution
Stabilizing a Fully Zero-Confirmation Autonomous Mode
Built around an autonomous mode that never asks the user for confirmation, reporting only when it encounters an error or unexpected event. The top priority was completing everything from theme selection to upload with no human involvement.
04 // IMPACT & RESULTS
Results
Full HD
1920x1080 / 30fps output
3 accounts
Parallel YouTube uploads supported
Fully automated
One theme line to scheduled YouTube upload