Deliverable #2 · 2026-07-29
3D model concepts — composing creatures from parts
Playable, not a screenshot. Drag to rotate. Use the on-screen buttons to cycle creatures (cauldron, basic humanoid, witchy humanoid, horse, pegasus, a small original creature, centaur) and gear (hats, weapons, shirt, shoes). “Surprise Me” randomises the combination. First load pulls ~40MB — it’s a raw engine export, not yet optimised for the web.
The problem this solves
Land of Lor lets players choose their own body shape, colour, and clothing — and separately, lets the game generate creatures out of the 64-element recipe system (see deliverable #1). Both of those are the same underlying problem: an unbounded number of possible combinations, built from a bounded set of parts.
The obvious way to build a 2D top-down game’s creatures is hand-drawn sprite sheets — a full set of directional walk/run/attack animations per creature. That approach works fine when the cast of characters is fixed. It breaks completely the moment players can choose their own gear, colour, and body shape, because the number of sprite sheets you’d need to hand-draw grows combinatorially with every new choice you offer. Four hat options × three weapon options × two shirt states is already 24 sprite sheets for one creature, before colour or body shape enter the picture at all.
The approach: parts + sockets, not pictures
The proof of concept above takes the opposite approach: every creature is a 3D scene built from a small number of primitive shapes, and every piece of gear is a separate scene that snaps onto a named attachment point — socket_head, socket_right_hand, socket_torso, socket_feet, socket_back — wherever that socket exists on whatever creature is currently active. The gear doesn’t need to know anything about the creature wearing it beyond “is there a socket with this name” — which is exactly the kind of loose, recomposable coupling the whole recipe system depends on (see the Horse → Pegasus pattern in deliverable #1’s linked docs, where adding one filled slot to an existing pattern produces a related-but-distinct thing).
The proof of concept goes one step further than a single creature wearing swappable gear: the Centaur scene is built by nesting a Horse scene and a Basic Humanoid scene inside each other and letting the socket system resolve gear placement through both layers of nesting automatically. That’s a direct, concrete demonstration of “deeper, composite objects” — the same idea the design docs describe in the abstract (an object made of objects made of objects) already running, today, in an engine.
What this proves, and what it doesn’t yet
Proven: the socket-attachment mechanic works, including through nested composition, and low-poly flat-shaded primitives look intentional rather than unfinished — a real style, not a placeholder.
Not yet attempted: driving the choice of shape, colour, and part from actual element/trigram data, the way the real recipe system will. Every creature and gear item above was hand-placed by a person choosing what looked good — the POC doesn’t yet read a hexagram and derive a shape from it. That’s the next experiment, tracked as its own open item under object-system-experiments.
It’s also worth being honest about where this sits relative to the finished game: this is a POC, built to test one mechanic in isolation. It answers “does part-composition work at all,” not “is this the final art direction.” Expect the real game’s models, and probably its whole visual style, to look different from this by launch.
Sources & artifacts
← related epic: object-system-experiments
Also available as raw markdown.