Loading
Loading trajectory…Raw trajectory ↓
A CLOTH SIMULATION PLAYGROUND
Same cloth.
Different dynamics.
See what changes when you change the solver.
Scrub real trajectories, compare parameters, and take the inputs with you.
01 / EXPLORE THE MOTION
A shared view of different solvers.
Drag either view to orbit both. Scroll to zoom. Cameras, scale, and playback time stay synchronized.
Loading
Loading trajectory…Raw trajectory ↓
Loading recorded cases…
02 / INSPECT THE PARAMETERS
The settings behind the motion.
Identical nominal material values do not imply identical constitutive models. Numerical settings differ by backend; ¹ Historical wall times use backend-specific settings and exclude some startup costs; they are not an equal-compute speed benchmark.
| Recorded parameter / metric | A | B |
|---|
Full parameter snapshots and failure gates
03 / KNOW WHAT YOU ARE SEEING
Inspectable results.
Explicit limits.
All ten clips use the same source garment, nominal medium material, and two-second horizon. A “passed” badge only means the historical sampled-frame checks passed. It does not establish real-cloth accuracy or solver superiority.
Failed PhysX cases stay visible, including self-intersection and motion-check failures. XPBD uses ClothForge contact extensions; the adapter is not a continuous collision detection guarantee.
The browser displays vertex positions rounded to 10⁻⁶ m. Original NPZ files, validation records, initialization records, and hashes are included. GPU runtimes were not re-run for this release.
Run a backend locally ↗