At Stability AI, we strive to advance the state-of-the-art in vision foundation models with diverse applications in understanding, reconstruction and generation tasks in both 2D and 3D. A central research question is how to train generative models that can act as generic yet powerful priors for the visual world around us.