In a recent paper, we presented the first semi-analytic model of galaxy formation in which the thermally pulsing asymptotic giant branch (TP-AGB) phase of stellar evolution has been fully implemented. Here, we address the comparison with observations, and show how the TP-AGB recipe affects the performance of the model in reproducing the colours and near-infrared luminosities of high-redshift galaxies. We find that the semi-analytic model with the TP-AGB better matches the colour–magnitude and colour–colour relations at z∼ 2, both for nearly passive and for star-forming galaxies. The model with TP-AGB produces star-forming galaxies with red V−K colours, thus revising the unique interpretation of high-redshift red objects as ‘red and dead’. We also show that without the TP-AGB the semi-analytic model fails at reproducing the observed colours, a situation that cannot be corrected by dust reddening. We also explore the effect of nebular emission on the predicted colour–magnitude relation of star-forming galaxies, to conclude that it does not play a significant role in reddening their colours, at least in the range of star-formation rates covered by the model. Finally, the rest-frame K-band luminosity function at z∼ 2.5 is more luminous by almost 1 mag. This indicates that the active galactic nuclei feedback recipe that is adopted to regulate the high-mass end of the luminosity function should be sophisticated to take the effect of the stellar populations into account at high redshifts.