TRIBE brain-prediction ad-test explainer — spoolcast pilot video
5-minute illustrated explainer on Meta's TRIBE brain-prediction AI tested on a real ad A/B campaign — first end-to-end output from the spoolcast pipeline.
First end-to-end output of the spoolcast pipeline. A 5-minute illustrated narrated video about Meta's TRIBE — a model that predicts brain-response patterns from video — and what happened when it was tested on a real ad A/B comparison (the model picked Video A, the market's ROAS picked Video B).
Every part of the video is AI-generated and the entire pipeline runs automated. No human illustrator, no voice actor, no manual editing. 44 hand-drawn-looking stick-figure scenes generated by kie.ai (nano-banana-2, locked to a single style anchor so every chunk matches), narration spoken by Google Chirp3-HD (Puck voice at 1.2x), reveal animations rendered programmatically by an OpenCV preprocessor (chalkboard eraser wipes for chapter boundaries, organic paint-style strokes for standalone scenes), and content-aware camera moves applied automatically to chunks where the rule engine decides the image needs to feel alive. Even the YouTube title, description, and thumbnail were AI-generated against the script. The human writes the narration script and reviews; the pipeline does the rest.
Single editable shot-list xlsx is the source of truth — every downstream stage (image gen, voice gen, frame preprocessing, Remotion headless render) reads from it. Edit one row, rerun one preprocessing step, rerender. The platform itself is tracked as a separate project; this entry is the first piece of content that came out of it.