Synthetic humans and creatures, fully sovereign (NishiLang, no third-party encoders): motion-on-demand rendered to browser-native animated PNG, and one recipe system that emits a whole bestiary via swap slots. Everything below was generated by the ecosystem and gate-verified today.
Text to motion (neural): a sentence -> a learned perceptron encoder -> motion class -> generated tokens -> the rig -> a fresh clip.
Talking avatar (lip-sync): a face that mouths your words (vowels open, plosives shut the lips).
Named move · and a swap-slot creature (choose species, head, ears, tail, build):
Coach - how's your form? An arm-raise, graded. Slide in an elbow-bend error; the coach scores your form and names the off joint (0 error = perfect).
A move is DATA (keyframes) -> 17-joint rig -> z-buffered render -> 16-frame APNG.
nx_motion_video_gate 4/4
The continuous smin-fused body posed by the same rig, rendered on the production skin shader (SSS, pores, spec, AO). 128x96 shown 3x - honest pixels.
nx_motion_photo_gate 4/4
A variant is a tuple: species x head x ears x tail x build. Row 1: human, horned, beastman (snout+tail), bulk. Row 2: elf (slim, pointy ears), goblin, horned goblin, snouted goblin. Row 3: hobgoblin, wolf, dire wolf (horned, bulk), coyote (slim). 12/12 tiles pairwise distinct; a head-slot-only swap visibly changes the render. nx_assetvariant_gate 4/4
The same tuples exported through surface nets as CLOSED, watertight meshes. STL (print-ready) and OBJ (with smooth per-vertex normals - imports clean into Blender / Unity / Godot):
STL: human - elf - goblin - hobgoblin - wolf - direwolf
OBJ: human - elf - goblin - hobgoblin - wolf - direwolf
glTF (.glb, web/engine/VR standard): human - elf - goblin - hobgoblin - wolf - direwolf
All six closed-manifold (Euler genus-0 band) and geometrically distinct - the swap slots change the MESH, not just the picture. Exports as STL (print), OBJ (+normals) and binary glTF (three.js / Unity / Godot / Blender / VRM foundation). nx_variant_mesh_gate 3/3 nx_obj_export_gate 2/2 nx_gltf_export_gate 4/4
The threads converge: one reusable synthetic body, diversified by SWAP SLOTS (horns + ears + tail), given MOTION on the same FK rig and shipped as video. The swap features are BONE-ANCHORED - the horns and ears ride the head as it turns. 43 parts, rendered on the production skin shader. nx_creature_motion_gate 4/4
And the ANIMALS move too: the bestiary wolf on an 18-joint QUADRUPED rig with a real diagonal walk gait (front-left strides with hind-right; knees bend mid-swing; tail sways). Same skeleton substrate, same shader, proven mechanically - opposing feet displace in opposite directions at peak phase. nx_wolf_walk_gate 4/4
| Render | Score /1000 | Verdict |
|---|---|---|
| Pioneer body, static, 512x384 (earlier arc) | 62 | CLAY |
| Pioneer body, static, 128x96 (same res as clips) | 218 | CLAY |
| Posed body mid-wave, 128x96 | 250 | CLAY - posing did not collapse quality |
The critic (min of fine-detail / structure / chroma, noise-hardened) calls PHOTOREAL at 700. Nothing here is photoreal yet; that is the point of having the critic.
| Capability | Industry SOTA (banked to disk) | Ours |
|---|---|---|
| Move library as data + want-a-move to video loop | Viggle, Kling, Runway, Pika | HAVE v0 |
| Part/species swap for asset diversity | (game pipelines; neural i2i for images) | HAVE v0 |
| Text-to-motion, free-form | MDM, T2M-GPT, MoMask | PARTIAL - parametric recipes |
| Pose-driven photoreal human video | Animate Anyone, MimicMotion, Champ | PARTIAL - clay-class SDF body |
| Face / clothes / hair swap on images | SimSwap, inswapper, IDM-VTON, Barbershop | GAP - S5 neural rung |
| Video diffusion backbone | SVD, AnimateDiff, HunyuanVideo | GAP |
Census (liar-killed): 156 permille at benchmark -> 406 after motion shipped -> 468 after the bestiary. 35/34 SOTA sources fetched live to knowledge/fetched/sm_*.raw over the sovereign TLS stack.
organs: nx_movelib, nx_figure_render, nx_bodypose, nx_apng, nx_assetvariant, nx_creature_motion, nx_quadruped, nx_critvis_lib | gates all GREEN: motion_video 4/4, motion_photo 4/4, assetvariant 4/4, variant_mesh 3/3, creature_motion 4/4, obj_export 2/2, wolf_walk 4/4 | tools registered: move_library, motion_video, asset_variant | sovereign nx_cc to nxasm_x86, no gcc, no 3rd-party encoders.