TryIto: Discrete IMEX Cloth Simulation with Learned Residual Forces

We propose TryIto, a hybrid cloth simulation framework that combines Kirchhoff–Love thin-shell mechanics with learned residual forces. The method introduces a time-varying (4D) latent UV atlas to improve texture-space continuity under large deformations, and a linearised IMEX time integrator that treats stiff elastic and damping terms implicitly whilst updating residual and external terms explicitly. To align optimisation with deployment, the framework minimises a discrete residual that mirrors the deployed numerical scheme, removing train–test drift typical of continuous-time PINN residuals under stiff time discretisations. We focus on the mathematical formulation, stability and conservation analysis, and computational complexity, positioning TryIto as a physics-informed foundation for high-fidelity, real-time garment simulation.