SUPER & ARCHER

For years my two daughters were Super and Archer. Super had a cape and could fly. Archer had a bow, which was a stick. The dog was in it too, mostly as an obstacle.

It went on long enough that it stopped being a game they played and turned into a place they both remember. They're at different colleges now. I wanted to build the place.

The older one draws them, and that turned out to be the hard part and the whole point. A drawing doesn't tell you where the cheekbone sits. It tells you what she notices.

What she notices is a solid dot for an eye, a head about as wide as it is tall, and hair that hangs instead of wrapping around the skull. I got all three wrong the first time and the faces looked like a stranger's. The heads have been rebuilt five or six times since.

Every character has its own build script now, and I rebuild one at a time. Working on all four at once meant all four stayed slightly wrong.

A Marin hillside, a house laid out like ours, eleven neighbours, and a forest at the far end that turns them into Super and Archer whether they were planning on it or not.

Super gets a jump you can hold past the top until the cape opens into a glide, and a sword that makes mailboxes wobble. Archer gets a bow that shoots where the camera looks and four kinds of arrow, one of which is a firecracker and one of which ropes her onto a roof.

Betsy the golden retriever sleeps under the lawn tree until somebody walks into the yard. Shoot an arrow into something nearby and she brings it back to your feet. Her walk, trot and gallop are real animation clips, blended by how fast she's actually moving. She was the single hardest thing in the project.

The last big piece was co-op. Two seats over Supabase, so both of them can be on the same hillside from two different colleges. It's the only feature in here with a deadline attached, and the deadline is that they don't live in the same house any more.

Three.js, ES modules, no build step. Characters are authored headlessly in Blender by Python scripts, so changing a face is changing code and I can rebuild all five models with one command.

The terrain is written once as a set of constants and read twice, by JavaScript for physics and by GLSL for the picture, so the ground you walk on and the ground you see can't drift apart. There's no engine under any of this, which was most of the fun.