Diffusion Models in Generative AI

摘要

Diffusion models have shown impressive capabilities in the generative AI space. These models have the capability to create images in a variety of styles from photorealistic and futuristic to many more artistic styles by simply using text prompts. This tutorial aims to introduce the underlying mechanisms that make these models successful along with hands-on exercises. The tutorial will start with explaining the diffusion concept with forward and reverse processes. Then, it will cover the fine-tuning process and the control procedures such as guidance and conditioning. The provided hands-on exercises will help apply these concepts on some real-world problems.