Options are essential tools in financial markets, and accurate pricing is critical. Traditional methods perform well but struggle with complex option features. Deep neural networks have been used to generate asset paths, but poor data quality often leads to inaccurate pricing. To address this, we propose a Monte Carlo option pricing method based on Diffusion Models (DM). Leveraging DM’s strong simulation capabilities, our method better handles complex features and time-series characteristics. We further enhance performance with a novel Decomposition-based LSTM-Diffusion Model (DM-L), which replaces the traditional path simulation step. Visual and quantitative evaluations confirm the high quality of the generated data. Experiments show that our method improves data quality by about 20% and pricing accuracy by about 15% compared to existing approaches.