← ClaudeAtlas

ardy-motion-generationlisted

Instalar y operar NVIDIA ARDY — modelo open-source de motion generation en tiempo real (autoregressive diffusion, control por texto/waypoints/teclado) — local en WSL2 sobre la RTX 3090, con el objetivo de generar locomoción de cuerpo completo (correr, saltar, vaultear, scene traversal) para el survivor de DayZ. Usar cuando el usuario pida "instala ARDY", "prueba ARDY", "genera animación con ARDY", "motion generation en tiempo real", "locomoción del survivor con AI", "animación por texto/waypoints", o retome el proyecto de locomoción DayZ vía IA generativa. NO usar esta skill para animación de armas/grip — está fuera de alcance por diseño, ver "Qué NO resuelve ARDY" más abajo.
willy92wins/DayZ-Modding-Knowledge-Pack · ★ 25 · AI & Automation · score 78
Install: claude install-skill willy92wins/DayZ-Modding-Knowledge-Pack
# ARDY — real-time motion generation (NVIDIA) ## What ARDY does NOT solve (read before investing time) The goal of this skill is **survivor full-body locomotion** (running, jumping, vaulting, scene traversal), NOT weapon animation. The actual weapon grip bottleneck in DayZ is **geometric parity** between the weapon model and the reference `.anm` on `Weapon_Root`/`RightHand_Dummy` (verified in-game, A6_SR2M project, 2026-06-17/23) — ARDY operates on a full-body skeleton (~27 "core" bones) with hand control as end-effector position, without any concept of held object geometry, individual fingers, or the DayZ ASI/IK system. Bringing ARDY to a weapon-grip problem would be solving the wrong layer. Full detail and comparison with 6 more tools: `<knowledge-notes>/ai-3d-pipeline/stage-05-animation.md`. ## What it is (verified against research.nvidia.com/labs/sil/projects/ardy/ and ## github.com/nv-tlabs/ardy — do not trust video/marketing, only primary source) **ARDY** = "Autoregressive Diffusion with Hybrid Representation for Interactive Human Motion Generation" (NVIDIA, research lab SIL). Two-stage autoregressive transformer denoiser: stage 1 predicts global root motion, stage 2 predicts body motion conditioned on root. Supports sparse kinematic constraints in time/joints. Real-time control: online text prompts, root trajectories/waypoints, full-body keyframes, end-effector joint positions/rotations, mouse waypoint editing, keyboard velocity commands, long-horizon goals. - Rep