Zenya model reused from the avatar loaded by JARVI3 Zenya Lab (vr_jarvi3_space.html).
Source: modules/metaverse_immersive_systems/app/avitars/zenya/zenya.fbx
Idle: relax-378947.fbx; talk: stand-talk-378997.fbx.
Web derivative: original geometry, rig and diffuse textures; textures capped at 1024.
Selected facial morphs are masked to head/eyelash material groups and split from the body using the Lab renderer's splitMorphRegion. Unsafe full-body morphs are removed.
The runtime uses the Lab's two-pass hair core/soft-edge fix and Z-axis jaw-bone opening. Upper-body framing, ActorCore arm gestures, head motion and blinks accompany timed lip shapes.
Speech: 75 fixed public utterances, en-US-AvaMultilingualNeural, matching the live Zenya Lab TTS voice. Local cached MP3 files and word-boundary mouth tracks; no visitor text sent for synthesis.
Three.js r134: MIT license, https://github.com/mrdoob/three.js/blob/r134/LICENSE

Portrait refinement: named root bones are animated once; ActorCore arm motion is limited to the body rig and its joint ancestors, matching the Lab. Original Top garment receives a 0.25-source-unit surface clearance in bind space and double-sided edge rendering. Hair uses a depth-writing core and a render-order-2 soft edge pass.

Live/TikTok motion refinement: head pitch-down 2.4 degrees, eye pitch-down 6 degrees; eye-level portrait camera. Original static/zenya-face-engine.js copied unchanged from JARVI3 source. Idle eye micro-movements and 9–18-second glances return to the lens on damped springs. 280 ms entry, 520–820 ms release hold and 1250–1950 ms spring-eased release; arm smoothing rate 9.5 / step 0.022 rad and authored weight 0.68 adapted to the portrait mix.
Alternate upper-body take: sit-talk_251075.fbx, SHA-256 274cd4e1428d31cc5186632c83e1008e030696fd3af6a4874b1b449809200da9; gzip animation JSON in zenya-talk-alt.bin, 16 arm tracks, Live time scale 0.86. The dashboard uses it on every fourth sufficiently long explanatory reply; short replies keep the standing take.
Lip sync follows interpolated prepared word cues on Audio.currentTime. Mouth lead is 0 ms for the direct browser audio path; the TikTok renderer’s OBS-specific 120 ms lead is not applied to this website. Buffering holds the current take; restart/resume never resets its loop.


Tailored outfit (2026-10-15): "Women's Blazer Black" by Latitude23.
https://sketchfab.com/3d-models/womens-blazer-black-43d65a4518c24b5c8e786657364b2c41
Licensed under Creative Commons Attribution 4.0:
https://creativecommons.org/licenses/by/4.0/
Adaptations: fitted to Zenya; closed front retained; sleeves redraped; collar and upper chest smoothed; rigged to Zenya's existing CC skeleton. Original 2K jacket PBR charts are retained. Normal vectors are compacted with 95th-percentile direction error below 2.5 degrees and stored as lossless WebP; other maps retain their source pixels. The full artist mesh is retained. Clothing attributes use sub-pixel web precision and compact typed-array storage; original face geometry and facial shapes are unchanged.
White blouse: custom Blender mesh with rolled collar, collar stands, placket, thread and four-hole buttons; woven cotton maps and contact shading baked from the real garment. Upper shoulder edges raised and fitted to prevent skin showing through.
The original face, hair, facial morphs, rig, voice and animation clips are retained. Covered body surfaces are masked beneath the outfit.