Tensor Network states (TNS) are effective representations of states describing low-dimensional complex many-body systems. In one-dimensional systems these variational states are successfully used in many algorithms, but in two dimensions it becomes difficult to exploit them to elaborate efficient algorithms.In this talk, a particular class of recently introduced isometric TNS (isoTNS) will be presented, through which successful numerical implementations are realized. As an interesting application, I will discuss recent results showing how to efficiently represent finite temperature states with isoTNS with low computational complexity.