We conclude the rigorous analysis of a previous paper concerning the relation between the (Euclidean) point-splitting approach and the local $\zeta$-function procedure to renormalize physical quantities at one-loop in (Euclidean) QFT in curved spacetime. The stress tensor is now considered in general $D$-dimensional closed manifolds for positive scalar operators $ \Delta + V(x)$. Results obtained in previous works (in the case D=4 and $V(x) =\xi R(x) + m^2$) are rigorously proven and generalized. It is also proven that, in static Euclidean manifolds, the method is compatible with Lorentzian-time analytic continuations. It is found that, for $D>1$, the result of the $\zeta$ function procedure is the same obtained from an improved version of the point-splitting method which uses a particular choice of the term $w_0(x,y)$ in the Hadamard expansion of the Green function. This point-splitting procedure works for any value of the field mass $m$. Furthermore, in the case D=4 and $V(x) = \xi R(x)+ m^2$, the given procedure generalizes the Euclidean version of Wald's improved point-splitting procedure. The found point-splitting method should work generally, also dropping the hypothesis of a closed manifold, and not depending on the $\zeta$-function procedure. This fact is checked in the Euclidean section of Minkowski spacetime for $A = -\Delta + m^2$ where the method gives rise to the correct stress tensor for $m^2 \geq 0$ automatically.