- RoleFull-stack engineer
- Year2026
- Built with
- React
- Vite
- Node.js
- Express
- PostgreSQL
- JWT
What was wrong
Video models answer a paragraph with a shot, so the paragraph carries everything: lens, lighting, motion, pacing, mood. Most people write two lines and get something unusable, then rewrite blind because nothing tells them which of the two lines was the problem.
What I did
The engine asks structured questions and assembles the blueprint from the answers, so nothing essential is left to chance. It also runs the other way: give it a video you like and it derives the prompt that would produce something similar, which is how most people actually learn what a good prompt contains.
What changed
The prompt engine is on its tenth revision, tracked separately from the application version because the two improve for different reasons.
What is inside
Structured questions, not a blank page
It asks about lens, lighting, motion, pacing and mood, then assembles the blueprint from the answers so nothing essential is dropped by accident.
Reverse-engineering from footage
Hand it a shot you liked and it derives the prompt that would produce something similar — the fastest way to learn what a good prompt contains.
A Genre DNA system
Eight specialised personas — medical, comedy, documentary, horror, marketing, cooking and more — each with its own visual language and pacing.
Sign in by email or Google
Work is saved to the account, so earlier blueprints stay available instead of dying with the tab.
From the build
A walkthrough of 1 screens, from the site and the dashboard.
Creating an accountSign up by email or Google. Work is saved to the account, so earlier blueprints stay available instead of dying with the tab.
Al-Baseera Centre
Still reading?
Then we should probably talk. Thirty minutes, free, no slide deck. Worst case you leave with a second opinion.