e3d1a0aa13
A 17-agent bench classified 103 models across the 125GB civit library and ran 12 live LoRA tests. Headline: NO LoRA BEATS THE PLAIN BASELINE — all 12 came back unusable, so the recommended recipe for both tiers contains zero LoRAs. The failure is at the checkpoint, not the LoRAs: comfyui_sd is the only farm operator that loads a LoRA and it can reach exactly one checkpoint, a person-photoreal SD1.5 merge that draws a woman for any clothing noun no matter how hard you negative-prompt it. flux_local, the lane that actually works, has no lora param at all. The two lanes are disjoint, so no LoRA here got a fair garment evaluation. Real bug found and fixed: the farm IGNORES ?parent_job=. A bogus id still returns the entire asset table (measured: 2,578 rows, 34 distinct parents), so mb_outputs()[0] was only ever correct because the table happens to come back newest-first — two jobs in flight would download each other's images, which is exactly what happened to two agents in the bench. Now filtered client-side. Measured changes: · Default backend CF -> flux_local klein-4b. At matched prompt and seed, klein holds corduroy wale across the garment where CF flux-1-schnell flattens it to smooth canvas. CF stays selectable: ~2s and never queues behind the farm's serial GPU lane. · Seeds are reproducible. /api/gen used time.time()%99991 so nothing could ever be reproduced, and /api/doll/gen inherited the operator default seed=42 so a phrase produced a byte-identical image forever and reroll did nothing. Both now hash the phrase + a variant counter; rerolling is bumping variant. · The painterly white doll2d template that actually shipped the 441 sprites wasn't in this repo at all — only djsim's later green variant. Added, and made the default; green is now the documented escape hatch for plaid and busy patterns, since the bench measured klein on pure white keying 7/7 flood seeds on both a dark solid and a busy multicolour knit. · Doll generation moved 1024 -> 704 to match the canvas, so we stop downscaling before the key and softening the edges the keyer depends on. · Templates hoisted to one registry; /api/gen returns the resolved prompt and seed and the UI shows them under the preview. · Local keyer warns loudly: flat3d product shots come back on studio grey with a drop shadow (border 215-244, zero pixels >=250), which only partially seeds the >232 gate. RMBG is semantic and unaffected. Verified: seeds deterministic and variant-sensitive, templates resolve view/year, and a live klein-4b generation produced a flat-lay corduroy jacket with legible wale and no person. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> |
||
|---|---|---|
| tools | ||
| web | ||
| .gitignore | ||
| blender_ops.py | ||
| README.md | ||
| server.py | ||
| STATE_AND_PLAN.md | ||
WARDROBEGOD — one skeleton, infinite fits
The wardrobe generator bench: load bodies (FBX/GLB/OBJ), build a clothes library, generate new garments on the farm, fit them, assemble dressed NPCs. Sibling of NPCFACTORY (the dress-up/clip bench) and MeshGod imagelab — this one owns clothing.
python3 server.py → http://localhost:8150
WG_HOST=0.0.0.0 python3 server.py → expose on the tailnet
MB_TOKEN=mbt_… python3 server.py → enables the generator tab (MODELBEAST bearer token)
No build step: stdlib server + three.js from CDN. Heavy lifting = local headless Blender
(blender_ops.py, tested on 5.1.2; override path with WG_BLENDER).
What it does
- Library panes: bodies (scans
library/bodies/+~/Documents/anatomy+ thriftgod'sweb/assets/models), garments, generated images, finished outfits. Upload via file inputs. - Stage: orbit viewer; FBX auto-converts to GLB (cached in
.glbcache/); HUD reads tris / height / rig state (bones + clips); first clip auto-plays. - Body ops: scale-to-height (bases arrive 1.0m tall — normalize to ~1.72m);
decimate-keep-weights (the SAFE local path for rigged meshes — never farm
/finish). - Rigid attach (tier 1): hats/bags/shoes → pick a bone, slide offsets, preview live.
Bake permanently via the fleet
/rigendpoint (m1ultra :8011) when happy. - Deforming FIT (tier 2): weight-transfers the rigged body's skinning onto a garment
mesh (Blender data_transfer, nearest-face interpolated, optional inflate-mm of air),
parents it to the armature.
FIT → wardrobe item= garment+skeleton GLB (reusable);fit → dressed glb= body+garment in one. - Assemble: body + ticked fitted garments → one dressed GLB in
out/— drop straight into thriftgod / djsim / procity, or publish to 3GOD.
Generating NEW clothing (the pipeline, $0 on MODELBEAST)
- flux (
flux_local, Klein) — product-shot prompt template: garment laid flat, white bg - cutout (
bg_remove_local) — the biggest quality lever before 3D - then ONE of:
- → rigid 3D (
trellis_mac, ~5 min): hats, shoes, bags, glasses — bone-parent these. NOT for floppy cloth: reconstruction eats thin geometry (fleet traps ledger). - → hanging texture: flat garment PNG for rack planes (djsim
fittings.js garment({image})) and the paper-doll wardrobe. Flat clothes love this route. - deforming wearables: model/retopo against the base body (or socket-cut a clothed torso, character_kit_modular style), then FIT. Never raw-gen these.
- → rigid 3D (
House rules
- Bases (anatomical models) stay in the library. Only dressed outfits ship into games.
- FIT requires a rigged body. Unrigged base → rig first (MIRPAMO on ultra, or Mixamo).
- One skeleton (mixamorig) across the fleet: any garment fitted on one body fits every body on that skeleton, and every clip in the bank drives it.