We explore the onset of star formation in the early universe, exploiting the observations of high-redshift LBGs and Lyα emitters (LAEs), in the framework of the galaxy formation scenario elaborated by Granato and coworkers, already successfully tested against the wealth of data on later evolutionary stages. Complementing the model with a simple, physically plausible recipe for the evolution of dust attenuation in metal-poor galaxies, we reproduce the LFs of LBGs and of LAEs at different redshifts. This recipe yields a much faster increase with galactic age of attenuation in more massive galaxies, endowed with higher SFRs. These objects have therefore shorter lifetimes in the LAE and LBG phases and are more easily detected in the dusty submillimeter-bright (SMB) phase. The short UV-bright lifetimes of massive objects strongly mitigate the effect of the fast increase of the massive halo density with decreasing redshift, thus accounting for the weaker evolution of the LBG LF, compared to that of the halo mass function, and the even weaker evolution between z~6 and z~3 of the LAE LF. The much lower fraction of LBGs hosting detectable nuclear activity, compared to SMB galaxies, comes out naturally from the evolutionary sequence yielded by the model, which features the coevolution of galaxies and active nuclei. In this framework LAEs are on the average expected to be younger, with lower stellar masses, more compact, and associated with less massive halos than LBGs. Finally, we show that the IGM can be completely reionized at redshift z~6-7 by massive stars shining in protogalactic spheroids with halo masses from a few times 1010 to a few times 1011 Msolar, showing up as faint LBGs with magnitude in the range -17<~M1350<~-20, without resorting to any special stellar IMF.
