Books, Generative AI
Book Review: A Damn Fine Stable Diffusion Book by Will Kurt
My thoughts after reading the first three chapters of this book by Manning Publications
I finally found the time to write an article on the blog…Recently, in fact, I have been busy with various activities that have entirely absorbed me, including completing the revision of my book, which has finally gone into production. Which means it should be released in August. But that’s another story. Today we talk about another book, which is really worth reading: A Damn Fine Stable Diffusion Book, by Will Kurt.
Before reading this book, I had no idea how Stable Diffusion worked, and to be honest, I didn’t even know very well how to frame it. After reading the first four chapters currently available in MEAP, everything suddenly became clear.
But let’s proceed with the order.
The book is about Stable Diffusion, a generative artificial intelligence model that produces photorealistic images. It introduces the topic progressively to facilitate the learning process.
It all starts with the simple creation of an image of a cup of coffee, and through more detailed tricks, you can produce stunning images of coffee cups.
Furthermore, the author underlines in the book that generative AI models are not random; the same sequence of images can always be generated if a particular trick is used.
Do you want to know what trick this is?
Read the book's first chapter, and you will know… but if you are really impatient, I can tell you that the trick to always generate the same sequence of images is to initialize the model with the same seed.
As you read, new concepts are added. While the Stable Diffusion model is seen as a black box in the first chapters, it is opened in Chapter 4, and the secrets of how it works are revealed.
My Thoughts
One particularly struck me as to why current Generative AI models for images are inaccurate in generating human faces. The explanation is that the models do not work on the initial image but on a very simplified version called latent (it goes from 400k pixels of the original image to 60 pixels of the latent image). This explains why details such as human faces are not rendered accurately when you want to go back from the latent to the original image.
The book also includes the Python code to run the model locally since it is open-source. That speaks volumes about the cost savings of using paid models like DALL-E.
The Book in Summary
Below, I summarize some features of the book:
- Author: Will Kurt
- Title: A Damn Fine Stable Diffusion Book
- Publisher: Manning Publications
- Number of Chapters: 17
- Who can read this book: Everyone with basic programming knowledge.
Do you find the book exciting and want to know more? Visit its official page!
Other books to add to your bookshelf…
Before leaving
I will be interviewing the author, Will Kurt, soon. If you have any particular questions, please leave a comment below.
Thanks, and see you next time!
