Option A2. The same rig, poses and procedural animation as option A, but rendered as ordinary
HTML: every slot is an <img> positioned with a CSS matrix() computed from the shared bone
math. No canvas, no WebGL — avatars are DOM nodes, so name tags, tooltips, hover states and CSS transitions come for free.
Raster layers are tinted with an SVG filter (flood × multiply); the accessory is an inline SVG whose fills are a CSS
variable — that is what a vector art pipeline gives for every part (crisp at any zoom, recolour = change a variable).
Watch the fps in room mode: 10 avatars × 15 layers = 150 transformed elements per frame. Fine on desktop; this is the option most sensitive to low-end phones.