← All work
Video automation · Marketing for NextUp
Automated Ad Pipeline
Instead of editing every ad by hand, I wrote a pipeline that builds them. You describe an ad in a JSON spec, and it produces a finished video in each aspect ratio, using the app's real theme renderer for the product shots.
- Role
- Solo
- Output
- 9:16, 1:1 and 4:5 video
- Stack
- Node · ffmpeg · sharp
- Used for
- NextUp ad campaigns
Why build it
Testing ads means making many variations: different hooks, themes, and formats. Editing each one by hand is slow, and it drifts from what the app actually looks like. The pipeline makes a new variant a config change.
Two generators
- Hybrid ads: AI-generated lifestyle footage (Kling, through Replicate, cached by prompt so re-runs are free), then a theme reveal rendered with the backend's own theme code, then hook text, an end card, and music, composited with ffmpeg.
- Carousel ads: a scripted 10-second sequence. A stock home screen transforms into a theme, then theme cards sweep across a curved "coverflow" arc to an end card. It's composited frame by frame with sharp.
Supporting tools
- Renders the Google Play feature graphic from the same source art.
- Cleans up raw screen recordings.
- A dry-run mode builds everything except the paid generation step, so specs can be checked for free.
The people in the hybrid ads are AI-generated footage. Every phone screen is rendered from the app's real themes.
Output
Ads it generated
Let's build something.
I'm open to full-time and contract roles in AI/agent engineering, mobile, and full-stack TypeScript, remote or hybrid around NYC and Philadelphia.