artlu.ai
100 days. 100 features.
Just AI and an internet connection.
day 113/100 · 100 shipped · 0 to go
← all projects
day 35 · apr 21, 2026 · launched

AI generated video - what is spoolcast?

spoolcast — the illustrated short-form video pipeline for builders. core message: getting attention can be a passive process now.

video · pipeline · dev tool · ai
site ↗ github ↗ ▶ demo

What this is
An 8-minute AI-generated explainer video about spoolcast — the pipeline that rendered the video. the pipeline is both the subject and the thing that made it.

What it covers
- cold open: the problem (you build things, marketing is a separate job)
- anatomy: 6 rules that keep the script from being slop
- budget: ~$3 per 5-minute video
- the four layers: image / animation / voice / render — what each does and how they connect
- proof: the actual pilot video this pipeline produced
- the agent layer (still being built) that turns raw signals into shot lists automatically

How it works
Script becomes a shot list — one row per sentence the narrator says. Each chunk gets one AI-generated illustration, style-locked via image-ref chaining. A python script computes a per-pixel reveal-time map for stroke-by-stroke paint-in. Google cloud chirp3-hd handles narration at 1.1x. Remotion renders headless into an mp4.

The 9-render iteration exposed the gap between "script works on paper" and "video lands on a viewer." Fixes along the way: a reveal-group concept for rhythmic beat clusters, act bumpers at section boundaries, a preview-must-explain rule for act openers, zero-prior-context rule for thumbnails, anti-self-hedge rule for title/thumbnail language, targeted speaking-rate adjustments on specific beats that landed too fast.

Built with
Remotion, React, Python, OpenCV, kie.ai (nano-banana-2), Google cloud chirp3-hd, Claude Code (editorial agent).

stack

ReactPythonspoolcastRemotionGoogle Cloud TTSkie.aiOpenCVClaude

media

loading repository tree...
← select a file to view