How to Read This Book

This book is a build manual, not a survey. It is meant to be worked through with a terminal open and a character bible half-filled on a second screen. Reading it end to end in one sitting is possible and largely useless.

Three ways through it

Pick one before you start. The parts are ordered so that a reader who follows the full path never meets a concept before it is grounded, but the shortcuts are honest — they tell you what you are skipping.

The seven-day start. Parts II → IV → V → VIII. You get one persona, one identity method, one production pipeline, and three distribution channels. You skip the market analysis, the law, and monetization. This path works, and it will make decisions you have to revisit: most people who take it train a LoRA before they have checked whether their chosen niche can pay for anything. Use it if you need to see a persona exist before you can commit.

The full build, ninety days. Every part in order, including the gates. The gates are one-question checkpoints — real decisions, not rhetorical ones. If you cannot answer the question, you are not ready for the next part, and skipping ahead will cost you more than the delay.

Scaling and operations. Parts XI → XII, for someone who already runs a persona and is considering a second one, a team, or an exit. It assumes Parts VIII and IX are behind you.

The shape of every chapter

Chapters are deliberately predictable. Once you notice the shape, you can navigate by it.

1. The problem, stated concretely. No history of the field, no opening about how fast things are moving. The pain comes first.
2. What you leave with. The artifact, named in one sentence.
3. The body. Numbers, thresholds, and the failure mode attached to each claim.
4. The failure modes. What breaks, what it looks like, what it costs to fix.
5. The artifact. The thing you now have on disk — a file, a setting, a checklist.
6. The gate, where a real decision exists. One question.

Fourteen parts end with a named artifact. If a part leaves you with nothing you can point at, the part failed, and that is a defect worth reporting to the publisher.

How claims are marked

This field is full of numbers that were true once, true for someone else, or never true. Four markers appear throughout, and they mean exactly this:

A chapter with no marker is either about method, or it is hiding something. When you meet a number with no marker and no date, be suspicious of it. That rule applies to this book first.

How money is written

USD everywhere, because every vendor bills in USD. Where a local currency helps, it appears in parentheses in that language's edition. Console prices are stated the way the vendor states them — monthly subscription, per second, per character, per thousand characters — because converting them all to one unit hides the traps. Chapter 7.5 is where the units are made comparable, and the outcome there is uncomfortable: the same image can cost $0.006 or $0.36 depending on how you buy it.

Time is money here and is counted. A technique that reaches 95% identity consistency but eats six hours a week of manual culling is presented as exactly that, not as a 95% technique.

Before you start

Three things, and it is better to discover a shortfall now than in chapter 4.5 with a dataset half-built.

• Time. Three to four hours a week for production, plus chat if you take the subscription path in Part IX. This is not passive income, and chapter 9.3 explains why the phrase is a lie in this business.
• Money. A solo stack runs $36–212 per month before you spend anything on ads. Chapter 0.4 gives you the arithmetic to decide which tier you can justify, and the decision rule is a cost per published post, not a feeling about the tools.
• A machine. If you plan to train identity models locally, 24 GB of VRAM is the realistic floor and 16 GB is not enough. Chapter 3.3 covers the thresholds and the cloud alternatives, and it is worth reading before you buy a GPU.

A note on the order

Parts III and IV come before Parts V to X on purpose, even though a reader might reasonably want the distribution chapter first. Identity is the constraint that governs everything downstream: your tool choices, your cost per post, whether you can move between vendors, and whether your persona survives a model being switched off. Get it wrong and every later part gets more expensive.

Part XI sits in the middle rather than at the end because the rules it covers are not housekeeping. Disclosure is a production step, not a legal afterthought, and a chapter written without knowing what has to be labeled is a chapter you will rewrite.