Deep generative models have transformed various domains, including drug discovery and development. Building on the exponential growth in computational power and extensive public datasets, these models facilitate the rapid design of novel molecules with targeted properties, possibly expediting the hit discovery phase. In brief, generative models are trained to learn patterns from extensive datasets, providing new structures as output. This chapter explores the potential and limitations of generative models in de novo design, covering key techniques, such as recurrent neural networks (RNNs), autoencoders (AEs), generative adversarial networks (GANs), and reinforcement learning (RL), focusing on their impact in pharmaceutical research. Despite recent successes, significant obstacles remain, including the reliance on high-quality datasets and the need for robust experimental validation, in order to bridge the gap between computational predictions and practical implementation. In fact, these models tend to generate compounds that may be chemically unstable or difficult to synthesize in the laboratory. The chapter also reports benchmarks, metrics, and performance evaluation methods through the examination of diverse case studies. Finally, the chapter discusses emerging directions, such as integrating quantum computing with generative models and leveraging large language models (LLMs), also emphasizing the need for cross-disciplinary collaboration and regulatory involvement.