The molecular design of peptide-assembled nanostructures relies on extensive knowledge pertaining to the relationship between conformational features of peptide constituents and their behavior regarding self-assembly, and characterizing the conformational details of peptides during their self-assembly is experimentally challenging. Here, we demonstrate that a hybrid-resolution modeling method can be employed to investigate the role that conformation plays during the assembly of terminally capped diphenylalanines (FF) through microsecond simulations of hundreds or thousands of peptides. Our simulations discovered tubular or vesicular nanostructures that were consistent with experimental observation while reproducing critical self-assembly concentration and secondary structure contents in the assemblies that were measured in our experiments. The atomic details provided by our method allowed us to uncover diverse FF conformations and conformation dependence of assembled nanostructures. We found that the assembled morphologies and the molecular packing of FFs in the observed assemblies are linked closely with side-chain angle and peptide bond orientation, respectively. Of various conformations accessible to soluble FFs, only a select few are compatible with the assembled morphologies in water. A conformation resembling a FF crystal, in particular, became predominant due to its ability to permit highly ordered and energetically favorable FF packing in aqueous assemblies. Strikingly, several conformations incompatible with the assemblies arose transiently as intermediates, facilitating key steps of the assembly process. The molecular rationale behind the role of these intermediate conformations were further explained. Collectively, the structural details reported here advance the understanding of the FF self-assembly mechanism, and our method shows promise for studying peptide-assembled nanostructures and their rational design.