Hi there! Thanks for reaching out to us on the VASP forum.
On the VASP wiki, we have a page dedicated to constructing Wannier orbitals using different methods. If you haven't checked it out already, I would recommend to start there. In general, creating Wannier orbitals can be quite tricky, as there is no systematic way to get what you want. The process often involves some degree of chemical intuition and trial and error.
If I understood you correctly, then you want to be able to generate atom-centered Wannier orbitals in VASP. This is possible with the LOCPROJ tag. You can specify what kind of orbitals you want to generate and on which atoms they should sit. These orbitals have the shape of atomic orbitals, including their symmetry properties (s, p, d, ...) and their centers coincide very nicely with the position of the atoms. The choice of k-point mesh depends largely on your system, as well as the size of the simulation cell. In order to obtain a good Wannier representation, it might be beneficial to include a slightly denser k-point mesh compared to your electronic convergence, but this is something you will have to test yourself.
If you want to improve the degree of localization, you can use Wannier90 via the LWANNIER90_RUN tag, for example. However, keep in mind that the MLWF procedure usually disregards the symmetries of the Wannier orbitals and can even move them around (we do not currently support symmetry-adapted Wannier functions in VASP + wannier90).
In general, metals are more difficult to Wannierize than more localized states/regions. This is mainly due to issues related to disentanglement, but to the best of my knowledge, there is also no formal proof that metallic states can be exponentially localized. We do offer disentanglement procedures in VASP that allow you to deal with metal-like regions in the electronic band structure (see CUTOFF_TYPE for more details), or you can use the iterative disentanglement procedure offered by Wannier90.
Finally, we also offer a more automatic Wannierization procedure via LSCDM. However, in this case the Wannier centers will most likely not coincide extremely well with the atom centers, and you have no control over which states are generated and in what order. Sometimes this can be beneficial, but in your case it might not be what you want.
