A whole human, generated procedurally, with every anatomical system present as an independently peelable layer -- skin, muscle, skeleton, and the digestive / circulatory / respiratory / urinary / nervous systems, ~60 named structures head-to-toe. Peel it like ZygoteBody, reshape it like Sloyd, and pose it on a real forward-kinematics rig (walk, sit, wave). 100% integer, sovereign -- no ML, no Blender, no CUDA. Rendered by our own integer ray-marcher.
Score is deliberately humbling on the SURFACE field (neural avatars, cloth sim, lip-sync) -- rung 1 of a ladder built from the hardware rung up. Where we already exceed: holistic internal anatomy (the field is surface-only), full sovereignty, hardware-up embodiment.
DIGITAL-HUMAN ADVERSARIAL CENSUS -- 2026-07-10 (them = banked awesome-digital-human field index; us = artifact-checked + execution-run organs) AXIS body-model (whole) PARTIAL critic: ours: a PARAMETRIC whole-body being (height/girth/frame/sex presets, gate-measured heavy>lean) with face+hands+internal systems; residual vs SMPL-X: LEARNED shape+pose blendshapes fit from thousands of scans + linear-blend skinning -> PARTIAL AXIS body reshaping HAVE critic: ours: parametric reshaping that MEASURABLY changes the silhouette (heavy 28318 > lean 20418 filled px, gate T3) across height/girth/frame/sex; residual vs shape-nets: reshape driven FROM a target photo -> HAVE AXIS avatar generation PARTIAL critic: ours: procedural full 3D human generated on a seed (persongen/bodyatlas), rendered sovereign; residual vs their text-to-3D / diffusion avatar: a trained generative avatar model -> PARTIAL AXIS head/face avatar PARTIAL critic: ours: INSIDE-OUT anatomical face (skull->muscle->fat->skin peel, nx_anatstack) + trained-neural skin; residual vs FLAME/DECA: learned expression blendshapes + detail displacement recovered from image capture -> PARTIAL AXIS hands (articulated) PARTIAL critic: ours: hands present as body regions; residual vs SMPL-X hands: 15-joint fully-articulated fingers with pose -> PARTIAL AXIS skin material PARTIAL critic: ours: trained-neural skin texture + family variation (nx_skin*); residual: measured subsurface scattering + pore-level micro-detail -> PARTIAL AXIS motion generation PARTIAL critic: ours: FK skeletal rig + a periodic, hierarchy-correct WALK CYCLE (nx_articulate, gate-proven); residual vs MotionDiffuse/MDM: text-conditioned DIVERSE motion synthesis from a learned motion prior -> PARTIAL AXIS gesture synthesis PARTIAL critic: ours: FK pose control can drive arms; residual: co-speech GESTURE generation conditioned on audio/text -> PARTIAL AXIS neural avatar render PARTIAL critic: ours: sovereign gaussian-splat renderer + anisotropic face splats + photo-fit (nx_gsplat); residual vs Gaussian Avatars: ANIMATABLE splats rigged to a body/expression model -> PARTIAL AXIS capture/recon PARTIAL critic: ours: gaussian photo-fit + 2-view SfM (cadtwin); residual vs ICON/PIFuHD: single-image IMPLICIT clothed-human recon with normal integration -> PARTIAL AXIS human image gen PARTIAL critic: ours: procedural person render + local GPU img2img finish; residual: a trained human-image DIFFUSION model -> PARTIAL AXIS clothed digitize GAP critic: ours: NONE -- we GENERATE nude anatomical beings, we do not reconstruct a CLOTHED person from a photo; theirs: PIFuHD pixel-aligned implicit hi-res -> GAP AXIS cloth simulation GAP critic: ours: NONE for garments (we have a WATER wave sim, not cloth dynamics); theirs: TailorNet/HOOD pose+shape-conditioned cloth drape -> GAP AXIS garment / clothing GAP critic: ours: the atlas being is unclothed; theirs: garment design + draping + dressing -> GAP AXIS speech / lip-sync GAP critic: ours: sovereign voice synthesis exists (voice-realism) but NO audio->viseme->face lip-sync driving a talking head; theirs: SpatialReal real-time lip-synced avatar -> GAP AXIS virtual try-on GAP critic: ours: NONE; theirs: warp/implicit garment transfer onto a body image -> GAP AXIS capture dataset GAP critic: ours: we GENERATE (could emit a labeled synthetic corpus) but publish no HUMAN capture dataset; theirs: THuman2.0/HuMMan real 4D scans -> GAP SCORE 294/1000 (HAVE 1 / PARTIAL 10 / GAP 6 of 17) OUR NAMED EXCEEDS (verified ABSENT from the banked digital-human field -- we are ahead here): - ★HOLISTIC INTERNAL ANATOMY: 8 anatomical SYSTEMS (skin/muscle/skeleton/digestive/circulatory/respiratory/urinary/nervous), 63 NAMED structures head-to-toe, INDEPENDENTLY PEELABLE (ZygoteBody-class) -- the awesome-digital-human field is SURFACE-only (clothing/motion/recon); no repo there generates a being with peelable internal organs (nx_bodyatlas, gate-proven) - SOVEREIGNTY: 100% integer, zero-dependency (no PyTorch/Blender/CUDA); theirs is a stack of trained nets on Python/CUDA - HARDWARE-UP: the being germinates on real silicon from CMOS/POST up (nishios spore), never-brick BY CONSTRUCTION -- an embodiment rung the field does not address - PARAMETRIC + PEELABLE IN ONE: Sloyd-style parametric proportions AND ZygoteBody-style layer peel unified in a single generated being THE LADDER TO SOTA -- FROM THE HARDWARE RUNG UP (operator's framing; each rung builds on the one below, gap-ordered within): R0 HARDWARE / EMBODIMENT -- being runs sovereign from CMOS/POST up, never-brick (nishios spore) .......... FOUNDATION (have) R1 BODY MODEL -- parametric whole-body (HAVE parametric; add learned shape+pose blendshapes + skinning) R2 ANATOMY SYSTEMS -- 8 peelable internal systems + named parts .................................. DONE (exceed) R3 HEAD / FACE -- inside-out face HAVE; add FLAME-class learned expression blendshapes R4 HANDS -- articulate 15-joint fingers on the existing arm rig R5 SKIN / MATERIAL -- neural skin HAVE; add measured subsurface + pore micro-detail R6 RIG / MOTION -- FK walk cycle HAVE; add text-conditioned motion (MotionDiffuse-class) R7 CLOTHING -- garments on the being (dressing the atlas) R8 CLOTH SIM -- pose-conditioned drape (TailorNet/HOOD-class) R9 HAIR -- strand-based hair/fur (fuzz displacement exists in anatstack) R10 SPEECH / LIP-SYNC -- audio->viseme->face on the head (voice-realism exists) R11 GESTURE / BEHAVIOR -- co-speech gesture + LLM-driven behaviour (our nofloat LLM) R12 NEURAL AVATAR RENDER -- animatable gaussian avatar rigged to the body model (gsplat exists) R13 CAPTURE / RECON -- single-image human recon (ICON/PIFuHD-class; SfM exists) R14 TRY-ON + GARMENT DESIGN-- virtual try-on + generative garment design NEXT (leverage-ordered, builds on what we have): R6 text-to-motion (rig exists) | R4 hands (skeleton exists) | R3 face expressions (anatstack exists) | R7 clothing (visible) | R10 lip-sync (voice exists) GAP/PARTIAL residuals feeding the ladder: - body-model (whole) :: ours: a PARAMETRIC whole-body being (height/girth/frame/sex presets, gate-measured heavy>lean) with face+hands+internal systems; residual vs SMPL-X: LEARNED shape+pose blendshapes fit from thousands of scans + linear-blend skinning -> PARTIAL - avatar generation :: ours: procedural full 3D human generated on a seed (persongen/bodyatlas), rendered sovereign; residual vs their text-to-3D / diffusion avatar: a trained generative avatar model -> PARTIAL - head/face avatar :: ours: INSIDE-OUT anatomical face (skull->muscle->fat->skin peel, nx_anatstack) + trained-neural skin; residual vs FLAME/DECA: learned expression blendshapes + detail displacement recovered from image capture -> PARTIAL - hands (articulated) :: ours: hands present as body regions; residual vs SMPL-X hands: 15-joint fully-articulated fingers with pose -> PARTIAL - skin material :: ours: trained-neural skin texture + family variation (nx_skin*); residual: measured subsurface scattering + pore-level micro-detail -> PARTIAL - motion generation :: ours: FK skeletal rig + a periodic, hierarchy-correct WALK CYCLE (nx_articulate, gate-proven); residual vs MotionDiffuse/MDM: text-conditioned DIVERSE motion synthesis from a learned motion prior -> PARTIAL - gesture synthesis :: ours: FK pose control can drive arms; residual: co-speech GESTURE generation conditioned on audio/text -> PARTIAL - neural avatar render :: ours: sovereign gaussian-splat renderer + anisotropic face splats + photo-fit (nx_gsplat); residual vs Gaussian Avatars: ANIMATABLE splats rigged to a body/expression model -> PARTIAL - capture/recon :: ours: gaussian photo-fit + 2-view SfM (cadtwin); residual vs ICON/PIFuHD: single-image IMPLICIT clothed-human recon with normal integration -> PARTIAL - human image gen :: ours: procedural person render + local GPU img2img finish; residual: a trained human-image DIFFUSION model -> PARTIAL - clothed digitize :: ours: NONE -- we GENERATE nude anatomical beings, we do not reconstruct a CLOTHED person from a photo; theirs: PIFuHD pixel-aligned implicit hi-res -> GAP - cloth simulation :: ours: NONE for garments (we have a WATER wave sim, not cloth dynamics); theirs: TailorNet/HOOD pose+shape-conditioned cloth drape -> GAP - garment / clothing :: ours: the atlas being is unclothed; theirs: garment design + draping + dressing -> GAP - speech / lip-sync :: ours: sovereign voice synthesis exists (voice-realism) but NO audio->viseme->face lip-sync driving a talking head; theirs: SpatialReal real-time lip-synced avatar -> GAP - virtual try-on :: ours: NONE; theirs: warp/implicit garment transfer onto a body image -> GAP - capture dataset :: ours: we GENERATE (could emit a labeled synthetic corpus) but publish no HUMAN capture dataset; theirs: THuman2.0/HuMMan real 4D scans -> GAP
Full-resolution peel filmstrip: nx_bodyatlas.png . sprite sheet: nx_bodyatlas_sheet.png
