Image processing · NextUp backend
Icon-Theming Engine
Every app icon is different: flat logos, photos, thin line art, text. The engine behind NextUp analyzes each one and re-renders it into a theme so a whole home screen looks designed, not decorated.
- Role
- Solo
- Runs on
- Firebase Cloud Functions
- Technique
- Classical computer vision
- Feeds
- NextUp Launcher
The problem
An icon pack only covers the apps its designer drew by hand. NextUp needed to theme every app on a user's phone, including ones no one has ever seen, so the styling has to be computed from the original icon.
How it works
- Structure extraction: Canny edge detection and Euclidean distance transforms find the shapes that make an icon recognizable. Text regions get detected and handled separately.
- Skeleton rendering: each theme redraws that structure in its own material, such as neon tube, carved stone, gold foil, or blueprint line.
- Delivery: rendering runs on Cloud Functions. The phone caches results on disk and shows adaptive previews while themes load.
The hard cases
The generic path breaks on icons that are photos, faces, or detailed illustrations. I built a comparison loop for those: the original, the previous output, and the fixed output side by side. I ran the iterations through my own agent dashboard, AgenticBrain, until each problem icon read clearly in the theme.
Results
Before, after, and every effect
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.