A strategy is presented for variational orbital optimization in time-independent calculations of excited electronic states. The approach involves minimizing the energy while constraining the degrees of freedom corresponding to negative curvature on the electronic energy surface, followed by fully unconstrained optimization, thereby converging on a saddle point. Both steps of this freeze-and-release strategy are carried out via direct orbital optimization at a similar cost as ground state calculations. The approach is applied in orbital optimized density functional calculations and is shown to converge intramolecular charge transfer excited states where the common maximum overlap method is unable to prevent collapse to unphysical, charge-delocalized solutions. The constrained minimization can also be used to improve the estimate of the saddle point order of the target excited state solution, which is required as input for generalized mode following methods. Calculations with the local density approximation and the generalized gradient approximation functionals PBE and BLYP are carried out for a large set of charge transfer excitations in organic molecules using both direct optimization as well as the linear-response time-dependent density functional theory (TD-DFT) method. The time-independent approach is fully variational and provides a relaxed excited state electron density that can be used to quantify the extent of charge transfer. The TD-DFT calculations are found to generally overestimate the charge transfer distance compared to the orbital optimized calculations, even when the TD-DFT relaxed density is used. Furthermore, the orbital optimized calculations yield more accurate excitation energy values relative to the theoretical best estimates for the medium and long-range charge transfer distances, where the errors of TD-DFT are as large as 2 eV.