A technique for physically-based cloth simulation uses linear upsampling operators. The upsampling operators enrich the appearance of a coarse mesh physical cloth simulation. The technique starts by pre-computing the upsampling operators using a pair of coarse and fine training simulations aligned with tracking constraints using harmonic test functions. Then the upsampling operators are trained using a novel regularization technique that enables mid-scale detail learning without over-fitting. Oscillatory modes may be introduced to add dynamic details not captured by the coarse mesh simulation alone. Trained upsampling operators can then be advantageously applied to coarse mesh simulations of cloth to add realistic detail to the cloth in real-time three-dimensional applications.