Observations of star forming regions in the solar neighbourhood reveal a surprising regularity, in the filamentary structure, the extinction and the initial mass function for instance. When considering more extreme galactic and cosmological environments however, this regularity does not appear. One possible explanation lies in the observational resolution that would prevent us to probe the scale of universality in a range of environments. The other possibility is that the large scale physics does imprint the sub-parsec scales and that star formation appear to be universal only in the narrow range of physical conditions where we do resolve it. In this talk, I will use high resolution galaxy and cosmology simulations to probe a series of physical conditions (Milky Way, interacting galaxies, high redshit) and to explore how these differences alter star formation. I will propose theories on the enhancement of star formation in mergers, and on the formation of massive star clusters, possibly being the progenitors of globular clusters.