I built this because the thing I care most about, generative image, video, and audio models, is moving faster than anyone can keep up with, and most of what is written about it is either too shallow to be useful or too dense to get through. I wanted one place that maps the entire ecosystem, every model, every company, every layer of the stack, and actually makes it understandable, so you can hold the whole board in your head instead of a scattered handful of pieces.
So the whole book is built around making something genuinely technical feel approachable, and almost every page is doing some quiet work to help. Hard ideas arrive with an analogy before the mechanics, so they land before they get precise, and where a picture explains more than a paragraph there is a diagram drawn like an instrument panel rather than a stock illustration. Key insights are pulled out on their own, because the point of a chapter is usually one non-obvious idea and it should not be buried. Inside scoops tell you what is really going on behind a model or a company, honest summaries admit what still does not work, and every term in the glossary is a tap away, defined right where you read it, so you are never stuck on a word.
And it is built to be learned, not just read. Each chapter opens with a short TL;DR so you can take the gist in one breath before deciding how deep to go, and ends with a quick check for understanding, because you do not truly learn something dense by reading it once. You learn it by being asked, and finding that you can answer. Your progress, your streak, and the questions you flag to revisit are all remembered, and they can follow you across devices, so the book behaves less like a document and more like a practice you return to.
How to read it. The book runs in two movements. The Machine explains how the models actually work, the foundations first, then the image, video, and audio models company by company, the craft of generation, and how to judge quality. The Medium turns to what gets built on them, the product stack and the ecosystem, AI filmmaking, the operator playbook, and where the field is headed. Read it start to finish and it builds on itself, or use the contents and the glossary to dip in anywhere.
How it is kept honest. It combines a deep grounding in the science and history of these models with live research that verifies every specific claim, a date, a version, a number, a ruling, against a primary or reputable source before it ships. Nothing is fabricated, sources are cited inline and listed in full, and a weekly pipeline keeps it current as the field moves. When something is uncertain, it says so.
The bet underneath all of it is that breadth and depth do not have to come at the cost of clarity. This is meant to be complete and thorough, the machine and the medium, how the models work and what we build with them, and still cohere as one readable whole rather than a pile of facts. If it takes something that felt impossibly technical and makes it feel learnable, and gets one more person from watching this happen to making things with it, it will have done its job.
Alex Park, Automata Labs.