Tutorial on implementing Diffusion Models

The following Jupyter notebook contains a tutorial on implementing diffusion models from scratch. Diffusion models are the state-of-the-art image generation algorithm and the backbone of Stable Diffusion, Midjourney or Imagen. We will see how the discrete-time formulation can be derived and how one can derive the continuous-time formulation from it. If you are more interested in the continuous time formulation as well as the generalization properties of diffusion models, please see my other notebook.