We present a novel method to generate quad meshes for non-rigid objects. Our method takes into account the geometry of a collection of key poses in one-to-one correspondence or even an entire animation sequence. From this input, on a common computational domain, an extremal metric is computed that captures the local worst case behavior in terms of distortion as the object undergoes deformation. An anisotropic, non-uniformly sized quad mesh is then generated based on this metric. This mesh avoids undersampling when deformed into any of the poses specified in the input and thus reduces artifacts. Hence, in contrast to previous approaches which target static geometry, our method aims to optimize the mesh’s adaptation to the shape for every pose expected during animation or deformation rather than for one specific reference state.