Code-first video & image, rendered natively on WebGPU.
28 — Physics dancer
A 6.6-second shot that puts most of the engine in one frame: glass word cards are thrown in from both sides and bounce off a dancer's raised arms and head, the last one landing on her hands; a cloud of embers parts around her silk dress. Renders in about 20 seconds.
pnpm fonts # Syne from google/fonts → assets/fonts/
pnpm render sim # analyses the clip and steps the physics without drawing (~35 s cold, ~6 s cached)
pnpm render # output/physics-dancer.mp4
pnpm render preview # live player; vision and physics fill in as it plays
pnpm render:src # same, on the engine's source (no package builds)
The clip is example 19's assets/dancer.mp4; generate it there with pnpm assets.
How it works
| Piece | API |
|---|---|
| Pose and silhouette, frame by frame as it renders | comp.vision(clip).person, placed with .place({ width, height, fit }); results persist in ~/.cache/framefields/vision |
| 3D physics that steps with the render | Physics.world3d(…) + world.live(comp): before each frame draws, its vision is ready and the world advances to it |
| Glass cards | Layer3D.roundedBox({ material: "glass" }) as the rigid body, the word as unlit blue Layer3D.extrudedText following the body's x/y/z signals just in front of it; lockRotation: true keeps every word upright |
| Thrown in from the sides | each card's launch velocity solved from gravity and a 0.7 s flight so the arc crosses her head and arms; the finale drops from above with a lower gravityScale |
| Arms and head that catch them | kinematic spheres along shoulder → elbow → wrist (and past it for the hand), following person.point(name, frame); without a camera layer the 3D z = 0 plane matches the video 1:1 |
| Embers | Layer.particles with push: { person } (her silhouette, hardest where it sweeps in), curl noise, colorBySpeed |
| Embers behind her | a second copy of the clip, cut out with .withVision({ mode: "matte" }), above the embers |