Recently, the realm of large-scale generative models spanning the image, audio, and many other related domains has made remarkable advancements. These strides enable the creation of samples that are, sometimes, nearly indistinguishable from real ones. Despite these significant advances, the challenge of exerting precise control beyond abstract instructions over what these models generate persists. This project aims to further our understanding of the internal mechanisms underpinning these models, harness this knowledge to enhance them, and further the control we can exert on what they generate. Ultimately, this project anticipates contributing novel approaches for designing and utilizing existing large-scale generative models that are more conducive to fine-grained control or even unlocking new use cases.