An automated pipeline that turns a researched business-history story into a finished vertical short: narration, word-timed captions, illustrated scenes, five visual formats.

A faceless story channel needs a video a day, and each one costs hours of narration, caption timing, art and layout. The pipeline renders one from a single story JSON, and the research tooling picks topics by what already over-performs.
Scans a set of channels, computes each channel's average views and surfaces the shorts running at five times that or more.
Classifies presentation, narration, captions, cut cadence and palette, and names the nearest house format.
A nine-slot arc is mapped onto an eight-scene structure and filled slot by slot from verified facts.
Captions are timed from TTS boundary events, so there is no transcription step and no drift.
Nine scene components and five full compositions (story, paper, dossier, terminal, blueprint) are chosen per story.
A shared style suffix and negative prompt per story make generated plates look like one illustrator.