Understanding How Stable Diffusion Works
In Chapter 4, we dove into the internal workings of the diffusion model with some math formulas. If you are not used to reading the formulas every day, it can be scary, but once you get familiar with those symbols and Greek letters, the benefit of fully understanding those formulas is huge. Math formulas and equations not only help us understand the core of the process in a precise and concise form, but they also enable us to read more papers and works from others.
While the original diffusion model is more like a proof of a concept, it shows the huge potential of the multi-step diffusion model compared with a one-pass neural network. However, some drawbacks come with the original diffusion model, denoising diffusion probabilistic models (DDPM) [1], and later Classifier Guidance denoising. Let me list two:
- To train a diffusion model with Classifier Guidance requires training a new classifier, and we can’t reuse a pre-trained...
