Diffusion modelsDiffusiondiffusion model, inspired by non-equilibrium thermodynamics, provide a powerful generative framework that progressively adds noise to data through a fixed forwardForward Markov process and then learns to reverse this noising process to generate realistic samples from pure noise. Diffusion modelsDiffusiondiffusion model can be conceptualized as an autoencoder: the forwardForward noising process acts like a fixed encoderEncoder, and the learned denoising network serves as a decoderDecoder that reconstructs data from noise. While this analogy is useful, diffusion modelsDiffusiondiffusion model are more generally understood as iterativeIterationiterative denoising or scoreScore-matching processes. This chapter presents a comprehensive introduction to diffusion modelsDiffusiondiffusion model, beginning with the foundationalDiffusiondenoising diffusion probabilistic model Denoising Diffusion Probabilistic ModelDenoisingdenoising diffusion probabilistic model (DDPM) and its various extensionsExtension. It further explores conditional diffusion models that enable controlled generation, as well as continuousContinuous formulations employing stochastic differentialDifferential equations for noise scheduling. Through this coverage, the chapter aims to equip readers with both theoretical understanding and practicalPracticepractical insights into the rapidly evolving field of diffusion-based generative modeling.