scnz — AI Video Director
Write a script; an AI director turns it into a finished video — visuals, voiceover, and an editable cut.
The role
Creator & Engineer — architecture & full build
Highlights
- Built an AI "director" that reads a script and turns it into a finished, watchable video — planning the whole thing so you start from a cut, not a blank timeline.
- Keeps a whole video visually consistent — characters, locations, and style stay coherent throughout, even as it is regenerated.
- Generates the visuals and an AI voiceover, with optional voice cloning — so a script becomes a watchable video without a camera or a mic.
- Runs on a team of AI agents for the different production tasks, led by the director — built on the same engine I open-sourced as agentside.
- A complete studio around it: a script editor and a timeline with transitions and audio mixing to refine the cut before export.
- Model-agnostic, so it is never tied to a single AI provider.
// inside_the_product
What scnz does — and what I built
AI Director
It turns your script into a finished video
Paste a script and the director plans and produces the whole video, making the creative calls a director would — so you start from a cut instead of a blank timeline.
Any Format
Made for every feed
Render the same video for YouTube, Reels, TikTok, or Shorts — landscape or vertical, sized for wherever it is going.
Voiceover
Narration in any voice
AI voiceover generated straight from your script, with optional voice cloning from a reference clip — so the whole video is voiced without touching a mic.
Timeline Editor
Edit on a real timeline
A zoomable timeline with transitions — cut, fade, slide, zoom — and audio tracks for music and SFX, so you refine the cut before you export.
// architecture
The system at a high level
A high-level view — a TanStack Start studio with two AI agents (a conversational director and a programmatic one) over a queue-backed generation pipeline.
- 01
Studio
ReactTanStack StartTiptapA TanStack Start app — a script editor and a zoomable timeline with transitions and audio mixing.
- 02
Agents
pi-agentTool-useSkillsA team of AI agents for the different production tasks, coordinated by the director that plans the video.
- 03
Generation
FALBullMQRedisImage and voiceover generation run through a background job queue, so the studio stays responsive while a video builds.
- 04
Data & storage
PostgreSQLDrizzleS3The project's content and generated media — structured data in Postgres, media in access-gated object storage.
- 05
Model layer
GeminiOpenAIpi-aiModel-agnostic LLMs across the pipeline, so no single provider is a lock-in.
What it is
scnz turns a script into a finished video. You write it — or have the AI draft it — and an AI director plans and produces the whole thing: the visuals, the voiceover, and an editable cut. It makes the creative calls a director would while you keep control.
I designed and built it end to end: the studio, the agents, and the generation pipeline.
The hard parts
- A director that reads a script. It plans the whole video and makes the creative calls a director would, so you start from a cut instead of a blank timeline.
- A consistent look. Characters, locations, and style stay coherent throughout, so the finished video looks like it was made together, not stitched from stock. Reference images and direction notes keep it on-brand.
- A team behind the director. A handful of AI agents handle the different parts of production, coordinated by the director — built on the same engine as agentside.
- A real studio. A script editor, AI voiceover with optional voice cloning, and a timeline with transitions and audio mixing — a full production pipeline, not a toy.
- Model-agnostic. Never tied to a single AI provider.
Want something like this built?