The dev behind the AI-heavy MMORPG Pawbound hit a combinatorial wall: with outfits, equipment, and cosmetics planned, characters baked into spritesheets would mean thousands of combination sheets. So he vibecoded his own tool — a JSON-driven 2D rig where parts separate, layers stack, and equipment swaps in real time, with no spritesheet generated per combination.
How the workflow works:
- Five Spine animations exported as JSON became the reference data: bone weights, mesh deformation, constraints, and timelines.
- He documented the Spine format, rig structure, JSON schema, and animation rules, then handed all of that context to Claude.
- Asking for “a walk cycle” or “an attack animation” makes the model generate poses and timelines against the existing skeleton; he reviews the result and nudges it on a timeline.
- Runs in the browser on Phaser with the Spine runtime.
His edit makes the thesis explicit: he is not claiming rigging is new — Spine and DragonBones already exist. What changed is the skill floor. Traditional tools demand studying animation principles, walk cycles, timing, weights, meshes, bones, and the software itself before producing anything usable. With this setup, the job becomes review instead of expertise.
This is the more interesting pattern of the moment: AI as a bridge into a discipline rather than a shortcut past it. The tool is game-specific and not public, but the workflow is portable — document a structured format, load the context, generate against it, correct the results. The comment thread carries the honest caveat: another dev spent six months attempting the same thing, and several readers are holding their skepticism until they can try it.