Nanometer-scale curvature patterns of an underlying substrate are imposed on lipid multibilayers with each pattern imparting distinctly different sorting dynamics to a metastable pixelation pattern of coexisting liquid ordered (L o)-liquid disordered (Ld) lipid phases. Therefore, this work provides pathways toward mechanical energy-based separations for analysis of biomembrane-associate species. The central design concept of the patterned sections of the silica substrate is a square lattice pattern of 100 nm projected radius poly(methyl methacrylate) (PMMA) hemispherical features formed by electron beam lithography which pixelates the coexisting phases in order to balance membrane bending and line energy. In one variation, we surround this pattern with three PMMA walls/fences 100 nm in height which substantially slows the loss of the high line energy pixelated Lo phase by altering the balance of two competing mechanism (Ostwald ripening vs. vesiculation). In another walled variation, we form a gradient of the spacing of the 100 nm features which forces partitioning of the Lo phase toward the end of the gradient with the most open (400 nm spacing) lattice pattern where a single vesicle could grow from the Lo phase. We show that two other variations distinctly impact the dynamics, demonstrating locally slowed loss of the high line energy pixelated Lo phase and spontaneous switching of the pixel location on the unit cell, respectively. Moreover, we show that the pixelation patterns can be regenerated and sharpened by a heating and cooling cycle. We argue that localized variations in the underlying curvature pattern have rather complex consequences because of the coupling and/or competition of dynamic processes to optimize mechanical energy such as lipid diffusion, vesiculation and growth, and phase/compositional partitioning.