Flagship Automation Project
YouTube
Video Maker
Bamboo Chronicles — a local, AI-assisted narration and auto-montage pipeline for long-form YouTube videos.
The project I’m proudest of. It takes the repetitive, error-prone work out of producing long-form videos — while every creative decision stays in human hands.
5
Production stages
80
Scenes demonstrated
80/80
Visuals mapped
34/34
Automated tests passing
The problem
Long-form video production is repetitive coordination work.
One narrative episode can hold dozens of scenes, prompts, image files, narration segments, subtitle timings and export requirements — and every manual step is a chance to introduce an error.
The solution
One local, five-stage workflow.
I designed a pipeline that validates the source packages, generates continuous narration, maps images to numbered scenes, builds an FFmpeg first cut, previews overlays, and exports a predictable delivery ZIP.
Scene & visual mapping
Keeping 80 scenes aligned.
The app detects scene numbers from different filename conventions (MK001.png, image_001.png, scene-014.jpg), converts imports to canonical project names, verifies coverage, and keeps each scene’s original narration and prompt accessible.
Narration automation
Continuous local AI voice.
Kokoro generates one continuous local narration instead of disconnected clips, then the app derives timing data and scene-aligned excerpts for montage and editing. (In the demo project the voice model is ready; that specific render is still pending — nothing is faked.)
Montage & overlays
FFmpeg first cut + auto-watermark.
FFmpeg assembles the first cut and supports selective rebuilds. Burned subtitles and an automatic watermark — opacity, size and position — can be adjusted and previewed locally before rendering the full video.
Editor handoff
One clean delivery ZIP.
The final checkpoint validates narration, timing data, mapped images, montage output and packaging before producing a single structured ZIP an editor can pick up without guesswork.
Reliability & validation
Tested where it matters.
Automated checks cover package roles, scene coverage, duplicate mappings, unsafe ZIP paths, canonical exports, subtitles, overlays and project reuse. The local suite currently passes 34 of 34 tests.
The machine handles
What I automated
Package validation, source/visuals detection, scene parsing, continuous AI narration, timing and audio excerpts, smart filename mapping, canonical organisation, coverage and duplicate checks, FFmpeg montage, subtitles and auto-watermark, selective rebuilds, editor-pack ZIP, and project status tracking.
I stay in control
What stays human
Scripts and finished images remain created by me or supplied by collaborators. The pipeline focuses on orchestration, validation, narration, montage and delivery — not automatic image generation. The story never gets lost in file management.
Built with
Technology & operations.
Built with
