0 Creator, Not Consumer

This chapter introduces the book’s central attitude: instead of passively consuming AI-generated content, readers are invited to become creators who understand, shape, and control the tools they use. It opens by criticizing the flood of low-quality AI output that can feel generic, repetitive, and disconnected from real human intention, then argues that generative AI becomes far more interesting when approached as a medium to experiment with rather than a shortcut to effortless output. The Grandmaster Flash analogy reinforces this idea: a tool originally meant for replaying existing material can become an instrument for invention when someone learns it deeply enough to transform it.

The chapter then explains the basics of generative AI in accessible terms, starting with what a model is and how it can be both a simplified mathematical representation of reality and a practical tool for prediction or simulation. It distinguishes discriminative models, which identify or classify things, from generative models, which can sample from learned patterns to produce new examples. Using handwritten digits as a familiar illustration, it shows that generative models are generally harder to build but open the door to far more creative applications. The chapter also notes that some models can function in both modes, depending on how they are used.

Finally, the chapter emphasizes the value of open models and open source tools for local image generation, especially Stable Diffusion and the surrounding ecosystem. It highlights advantages such as greater control, privacy, customization, community improvement, and freedom from dependence on commercial services. It also gives a realistic overview of hardware needs, explaining the difference between inference and training, and showing that many tasks are possible even on modest GPUs or Apple Silicon machines. By the end, the reader is promised a practical path toward producing increasingly impressive and personalized images at home, moving from basic experiments to highly customized creations that feel much more like authorship than consumption.

We use models everyday often without realizing it.

Sample of one of the most famous machine learning datasets

A simple implementation of a generative model for digits

Getting my imagination on the screen

Browsing an infinite library of Pulp Sci-Fi that never was.

Who knew monks were such avid readers of sci-fi?

Envisioning ancient aliens.

By the end of the book our results get very impressive

One of the most fun parts of working with diffusion models is customizing them

• Stable Diffusion is an open source model for image generation
• Using open models and open source tools give you full control over the process
• Discriminative models and answer questions about the world, generative models actually represent part of it.
• Diffusion models allow us to ‘remix’ from the collection of existing human generated images
• By using custom tools and building custom models we will be able to create some truly amazing images!

FAQ