Context. Surface brightness-colour relations (SBCRs) are largely used for general studies in stellar astrophysics and for determining extragalactic distances. Based on a careful selection of stars and a homogeneous methodology, it has been recently shown that the SBCR for late-type stars depends on the spectral type and luminosity class.Aims. Based on simulated spectra of late-type stars using MARCS model atmospheres, our aim is to analyse the effect of stellar fundamental parameters on the surface brightness. We also compare theoretical and recent empirical SBCRs.Methods. We used MARCS model atmospheres to compute spectra and obtain the surface brightness of stars. We first explored the parameter space of MARCS (i.e. effective temperature, log g, [Fe/H], microturbulence, and mass) in order to quantify their impact on the surface brightness. Then we considered a relation between the effective temperature and log g for late dwarfs and giants, as well as a solar metallicity, in order to allow a consistent comparison of theoretical and empirical SBCRs.Results. We find that the SBCR is not sensitive to the microturbulence and mass. The effect of metallicity on the SBCR is found to be larger for dwarfs than for giants. It is also larger when considering larger V - Ks values. We also find that a difference of 0.5 dex in metallicity between Galactic and LMC SBCRs does not affect the recent LMC distance determination, based on eclipsing binaries, by more than 0.4%. By comparing theoretical with empirical SBCRs, we find a good agreement of less than 2σ for F5-K7 dwarfs and giants stars, while a larger discrepancy is found for M dwarfs and giants (about 4-6σ). The surface gravity properties, as modelled in MARCS, explain the differences in the empirical SBCRs in terms of class. We finally find that theoretical and empirical SBCRs for Cepheids are consistent.Conclusions. Carefully considering metallicity and log g is mandatory when calibrating or using SBCRs.