It has been known for decades that present-day galaxies fall into two major classes: giant, red, passive ellipticals and more modest-sized, blue, star-forming spiral galaxies. It is likely that the present-day sizes and morphologies of galaxies are set by their stellar mass in the past, along with their star formation and merger histories and the environment in which they reside. However, the situation at high redshift is not well-understood, for a number of reasons: 1) Present-day classifications do not apply as readily to high-redshift galaxies, with a high proportion of clumpy, irregular galaxies and a substantial passive, disk-dominated population; 2) high-redshift galaxies are more compact on average than their local counterparts; 3) only the most massive high-redshift galaxies are typically measured; and, 4) optical high-redshift observations are in the rest-frame UV, which traces star formation well but not stellar mass. Consequently, it is unclear how the various masses and morphologies of high-redshift galaxies contribute to the morphologies we see today. In this talk I will describe our project using the new Gemini South Adaptive Optics Imager to achieve diffraction-limited, wide-field (85''), near-IR imaging of galaxies in clusters as distant as z~3. This imaging allows us to trace the morphological evolution of galaxies over cosmic time, and determine the contribution of stellar mass distribution to present-day morphology.