In essence, the story of generative models is one of balance—between tractability(computation) and flexibility (complexity). They are two conflicting objectives because tractable models, while easy to evaluate and efficient at fitting data, often struggle to capture complex structures in rich datasets. Conversely, flexible models excel at modeling intricate data patterns but come with high cost for evaluating, training or sampling. Diffusion model represents an advancement as it manages to combine these qualities, providing both analytical tractability with ability to handle complex data structures. However, despite of the attempts proposed to make the process much faster and more efficient, including LDM and other methods such as Improved DDPM by Nichol et al. in 2021, diffusion model still face challenges with efficiency and speed. This is primarily attributed to their reliance on long Markov Chain of diffusion steps or multiple forward passes for sample generation. This lag compared to its alternatives like GANs, highlights a critical area for future improvements.