AI video studio that turns one idea into a finished short video, ready for review
My own system that takes an idea, writes the script, adds a voice and pictures, edits it all together and checks it, ready for a person to review.
My own project. Built and tested, not a client job.
From idea to video
A production system built around review, quality and cost controls. Signal, an internal build.
- Step 1 Script Scene by scene
- Step 2 Voice Character audio
- Step 3 Scenes Visuals and motion
- Step 4 Edit Local assembly
- Step 5 Review A person decides
Built-in guardrails
- Budget set before anything is generated
- Quality checks on every stage
- Uploaded privately first
- 40 s
- A test video it made from start to finish
- Measured at 39.9 seconds, 1080 by 1920 pixels, 30 frames a second. It reached final review and was not published.
- Vertical
- Sized for Shorts, Reels and TikTok
- 1080 by 1920 pixels.
- Private
- Until a person approves it
- Videos upload privately, and a separate decision is needed before one can go public.
One connected production line
I designed and built a system that finds an idea, writes a script scene by scene, creates the voice and pictures, edits the clips together and produces a finished vertical video. A dashboard shows every stage, so a person can check the work as it goes.
The voices, images and video clips come from AI services the system connects to. The rest is my own build: the workflow, the editing and the controls.
Built to stay under control
It checks the budget before it spends anything. If one step fails, the finished work is kept and reused instead of being paid for twice. The storyboard and the final video are both checked, and a person can stop or approve the work along the way.
The test video made it all the way to final review. That proves the production line works, but not every possible route through it, and none of its videos have been published.
Technical notes
Built in TypeScript and React, with PostgreSQL and Redis behind the job queue and FFmpeg doing the final edit. Cost limits are checked before each paid generation call, and completed assets are kept so a failed stage can be retried without buying them again.
What this involved
Technical notes: what it is built with
TypeScript, React, PostgreSQL, Redis, FFmpeg, AI provider APIs.