Traditional, data-driven materials discovery involves screening chemical systems with machine learning algorithms and selecting candidates that excel in a target property. The number of screening candidates grows infinitely large as the fractional resolution of compositions and the number of included elements increase. The computational infeasibility and probability of overlooking a successful candidate grow likewise. Our approach takes inspiration from neural style transfer and shifts the optimization focus from model parameters to the fractions of each element in a composition. By leveraging a pretrained network with exceptional prediction accuracy (CrabNet) and writing a custom loss function to govern a vector consisting of element fractions, material compositions can be optimized such that a predicted property is maximized or minimized. It is expected that fractional optimization would excel in inverse design problems concerning the effects of dopants and those seeking a balance between competing properties.