Disclosure: This is a digital prototype. The physical bead piece has not been built yet.
I’m building Toonbead, a browser-based tool that turns photos into fuse-bead patterns. I wanted the first documented pet case to show the whole pipeline—including the part that failed.
Test setup
The source was a real gray-and-white cat photo. The workflow was:
- One protected Pet Icon AI restyle to simplify the fur and isolate the subject.
- Browser-local mapping to a 58 × 58 grid with up to 16 Perler Standard colors.
- Local counting and export: 803 beads, four 29 × 29 boards, a six-page PDF, PNG, and materials CSV.
The AI step prepared the image. Grid size, palette matching, cleanup, counts, and exports stayed local and did not use another generation.
The useful failure
I tested the one-click stray-bead cleanup. It removed several small white regions around the face. To the algorithm they looked isolated; to a human they were part of the cat’s muzzle and identity.
So I reverted the cleanup and kept the pre-cleanup mapping as the downloadable result.
That changed the product rule: cleanup should be reversible and visually reviewed, especially around eyes, muzzle, and markings. A lower bead count is not automatically a better pattern.
What the case includes
- original cat photo
- AI-assisted Pet Icon artwork
- transparent 58 × 58 bead-pattern PNG
- counted materials CSV
- six-page printable PDF with four board maps
Every asset is labeled as a digital prototype. There is no claim that a finished physical craft exists yet.
Inspect the full case and download the exact assets →
If you build creative tools, how would you handle this: keep cleanup off by default, or protect likely face regions before simplifying?
Top comments (0)