Deep neural networks, the cornerstone of the modern AI revolution, had their first breakthroughs in visual tasks (e.g., object recognition). This is not a coincidence given the high interest in such capabilities across industries (e.g., autonomous driving, medical imaging). However, to train, evaluate, and improve vision networks for specific applications, one needs extensive data. Transformative advances in generative AI (GenAI) have recently made it possible to synthesize highly realistic, complex visual data. Harnessing the power of synthetic data can overcome challenges faced with real-world data (e.g., scarcity, cost) and ultimately foster more visually capable machines. Here, we offer an overview of different types of visual GenAI and examine their adoption in real-world applications. We focus on autonomous driving (AD) and medical imaging applications because of their high societal impact and unique data challenges (e.g., privacy concerns for patient data, the long-tail challenge in AD). Finally, we discuss current limitations and promising future directions.