Files
animation/characters/rig-work
jeremy 0913a5972c chore(rig-work): archive GLM's LENA_rig_v1 exports, pulled from the game
The articulated-finger/straight-leg rig model GLM shipped into ariki-game
2026-08-15, replacing the sanctioned lena_leafbikini_base_v01 ship. Jeremy
pulled it from the game 2026-08-18; ariki-game bodies reverted to pre-swap.
The blend source does not survive — these exports are the only artifacts.
The LOD1 bake is the 'yellow Lena': one skin atlas wired into all three
texture slots, so metallic/roughness sample skin pixels (see README).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-18 11:31:10 -07:00
..

characters/rig-work — disposable AccuRig rig carriers

Decimated FBX copies of the unrigged Tripo bodies whose only job is to carry a skeleton out of AccuRig and back. AccuRig refuses the raw ~1.9M-tri FBX outright (confirmed 2026-08-04), so the bait exists purely to be small enough to rig.

Never ship a file from this folder. The graft takes geometry and materials from the pristine GLB and only the skeleton from the bait — see .agents/plans/rig-graft-lane-2026-08-04.md for the lane.

"Disposable" now has one exception, and it is about reproducibility, not shipping. A bait that fed a nearest-surface graft is a required input to that graft forever: it is the only thing index-exact with the AccuRig return that is also in the pristine GLB's space, so it is what recovers AccuRig's offset in closed form. Such a bait gets copied (not moved) into the ship folder alongside the AccuRig output. It is still never shipped to the game, and the copy here stays disposable. First case: lena_leafbikini_base_v01 — see that folder's README, "the bait is the bridge".

Per ../REGISTRY.md rule 11 this folder is exempt from the filename grammar and gets no Derived-artifacts rows; the grafted body's notes column names the bait its skeleton came from. The tool-era names are kept on purpose so these stay recognisable as throwaway.

Current contents (built 2026-08-04, moved here from ariki-game 2026-08-06)

file tris notes
lena_tripo_rigbait.fbx 77,999 from originals/female/female_lena_tripo.glb
lena_leafbikini_tripo_rigbait.fbx 77,999 from originals/female/female_lena_leafbikini_tripo.glb (2026-08-13); leaf-bikini garment sculpted on — fine for AccuRig (silhouette only), fingers distinct. Rigged and shipped 2026-08-13 as lena_leafbikini_base_v01; copy preserved in that ship folder
mako_tripo_rigbait.fbx 78,000 from originals/male/male_base_bald_tripo_v1.glb
*_qa_front.png, *_qa_hand.png QA renders; fingers fully distinct, no webbing

Built with tools/rigbait_decimate.py: region budgets body 24k / head 14k / hands 40k, no scale and no recenter (native ~0.98 m, feet at Z=0).

These two have hands. The lane is being rebuilt with the hands cut and a GLB sibling emitted from the same decimation; when that lands, these are superseded.

Traps

  • Baits are geometry-only (~3.5 MB), so AccuRig shows a grey model. That's fine for rigging, but it means the July "Dummy001 export" check can't use textures: verify AccuRig's output by mesh name + tri count instead. Correction 2026-08-13: the parenthetical used to read "a GLB's packed textures don't survive the FBX embed". They do — rigbait_decimate.py exports with embed_textures=True, and lena_leafbikini_tripo_rigbait.fbx carries 4K basecolor and normal in its .fbm/. What is actually true is the reverse direction: AccuRig strips them, and its return has no images at all. Verify by mesh name + tri count either way.
  • AccuRig exports come back to this folder as <character>_tripo_rigged.fbx (plus a settings .json and an often-empty .fbm/). The older _tripo_accurig.fbx spelling in the lane plan was never what the tool actually wrote.
  • These files used to live in ariki-game/assets/models/characters/rig-work/ and needed a .gdignore to stop Godot importing them. Nothing here is a game asset; that is precisely why the folder moved.