AI automation
A video pipeline as code
The system we edit our YouTube and Instagram videos with: research, script, storyboard, edit, and package, all driven from one repo with a code-based motion engine.
The problem
Content dies from friction. Every video meant juggling research tabs, script docs, an editing timeline, and packaging checklists across five different tools.
The result was slower output and a look that drifted from video to video.
What we built
The pipeline runs end to end: niche research over real channel data, scripting, storyboarding, then editing through an HTML/CSS motion engine, with packaging at the end. A talking-head mode handles real camera footage: transcript rough-cuts, then graphics layered over the top.
It already edits the videos on our channels, and the whole system installs into any repo as a plugin.
The result
The edit lives in a repo, not a timeline
2
editing modes: motion graphics and talking head
2
channels run from one repo
1
repo to run it all from