genbox

From wiki.gromacs.org

Used in the preparation of the coordinate file for the system. genbox can do one of 3 things:

generate a box of solvent.
solvate a solute configuration, eg. a protein, in a bath of solvent molecules.
insert a number of extra molecules at random positions.

[edit] Excluding regions

Sometimes you have a pre-generated region of structure inside which you don't want genbox to place waters, because they wouldn't be there in an equilibrated system and waiting for equilibration in simulation would take too long. There are several reasonable strategies here, if you observe genbox is placing waters where you don't want them. Below, "lipid" refers to this region, even if it's protein or something else.

genbox just uses van der Waals' radii for different atom types to generate volumes that exclude waters. If your lipid has interstices that are large enough to admit waters, then genbox will place them there.

Probably, an energy minimization of the lipid in vacuo will see it contract so that the molecules are pretty much close-packed, and hopefully this density will also be close to that for the subsequent simulation.
Alternatively, or if there are still holes that admit waters, you can make the radii of your lipid C and H atoms larger for the purposes of genbox using a modified gromacs/share/top/vdwradii.dat file. Be careful later with the gentleness of your energy minimization and equilibration schemes else you'll generate large structural changes and thus bad contacts and LINCS errors, etc.
Conversely, you could increase the radii of your water atoms, but be aware this will also affect behaviour at the water-lipid interface. This can cause water density away from the micelle to be too low, so a work around of this is to copy the vdwradii.dat file to current working directory then add entries for water so that it's radius is not increase i.e. for SPC add the following to three separate lines: "??? OW 0.105", "??? HW1 0.04" and "??? HW2 0.04". Then use the -vdwd flag with a value larger than 0.105. This way your water appears normally to the new waters but the micelle atoms all seem larger. Keep repeating with -vdwd values that are slowly getting larger until you find the minimal value of -vdwd for which no waters are placed inside the micelle.
You could use a molecule visualization program, find the waters inside your region and delete them (you could script this with some such programs, too).
If your lipid volume is regular (e.g. a spherical micelle) you could write a script to delete all water molecules within that volume (e.g. within an appropriate distance from the micelle center of mass)