Artificial intelligence (AI) has become a cornerstone in enhancing the efficiency and safety of oil drilling operations. It supports a variety of functions, including drilling status recognition, predictive maintenance, and risk assessment. Despite its advancements, AI in oil drilling faces a significant challenge due to the prevalence of long-tailed, imbalanced datasets. Critical scenarios such as pulling out, reaming, and high-risk operations are often underrepresented in available data, limiting the effectiveness of AI models. Addressing this data scarcity, our study introduces a novel approach using a diffusion model-based method to generate realistic and balanced synthetic datasets for drilling AI systems. This method not only provides a cost-effective alternative to acquiring real or simulated data but also enhances model reliability. Furthermore, we introduce new evaluation metrics specifically designed to assess the quality of generated data, demonstrating that our synthetic distributions closely mirror actual conditions. Our research paves the way for broader and more dependable applications of AI in oil drilling, promising substantial improvements in operational outcomes.