A key idea from Kisin's work on crystalline and semistable deformation rings involves constructing resolutions of these rings via moduli of Breuil-Kisin modules. For crystalline deformations with Hodge-Tate weights 0 or 1 the geometry of these resolutions closely models that of the deformation rings themselves, but for higher weights they are too large. I will explain a refinement of this approach which can be used to prove, for unramified extensions of Qp, potential diagonalisability of crystalline representations with weights ≤p.