This answer depends on how different the ligand binding sites are between the homology models, but you could align the homology models in pymol, and then copy the ligand from one of the other models into the structure of the ligand-less model.
If that’s impossible, you can try some complicated manual modelling with hybridize by frankensteining parts of the structure you want onto a structure that has a ligand. but that can be quite complicated and time consuming, and i’m not aware of any publications or protocols that do that without extra information like electron density. But you could be the first if it works!
.
