Free · Unlocks instantly on this page
AI video that doesn't look AI.
One block you paste into Claude or ChatGPT once. After that, hand it any product and a format — it writes the image prompt, the video prompt and the script to the rules we render with in production.
Six-part image skeleton
Subject, distance, camera, light, set, bans — the order and the exact capture stack we render with.
Per-format camera values
Testimonial, GRWM, podcast, street — phone position, framing and eyeline read frame-by-frame off clips that performed.
The words that break renders
"Photorealistic" paints. "Selfie" renders a phone. Seven phrases banned outright, each with its known failure.
Script rules included
Arguable stance inside two sentences, one escalating move, kitchen-table verbs, end on a completed beat.
Every rule in it came out of renders that failed first — face-filling crops, rainbow bookshelves, band tees with four printed faces. The prompt exists so you skip that year.
Want to check its work first? The five docs it's compiled from are free and ungated.