DescriptionPixel art remains a contemporary art form and a common rendering technique in digital games and media. However, the manual creation of pixel art is often time consuming and requires a degree of skill that is not easily obtained by novices of the art. Few, if any, methods exist to automatically generate pixel art. Naive downsampling techniques such as nearest neighbor and cubic downsampling do not adequately preserve features or maintain a vibrant palette. In this thesis we present our work on automatically and semi-automatically converting high resolution images into an output that approximates the manual results of pixel artists. This is a multi-step, iterative algorithm that simultaneously solves for a palette and a mapping of segments of the input image to pixels in the output. We provide a set of controls that give the user flexible influence on the output and the ability to work anywhere between a purely automated and purely manual process. We present the automated and semi-automated results of our algorithm and compare them to the results generated using naive downsampling techniques and the manual results produced by expert pixel artists. Through a formal user study and interviews with expert pixel artists, we demonstrate that our results offer an improvement over the naive methods.