Learning path

Understand the machine

A smart non-engineer who wants the whole mechanism, in order, without equations.

13 steps, in order

Denoising concept · step 1
The core loop everything else hangs off.
Working compressed concept · step 2
Why models do not operate on pixels.
In and out of pixels concept · step 3
The encoder and decoder, and the artifacts they own.
How words steer it concept · step 4
What a prompt actually does inside the model.
Attention and patches concept · step 5
The architecture shift that made scaling work.
Adding time concept · step 6
Where the cost and the difficulty come from.
Flicker and drift concept · step 7
The characteristic failures of video generation.
Steps and guidance concept · step 8
The two dials people misuse most.
Control surfaces concept · step 9
Everything you can hand the model besides words.
Teaching it your subject concept · step 10
How a specific character, style or look becomes repeatable.
Determinism concept · step 11
What a seed does and does not guarantee.
Known limits concept · step 12
Hands, text, physics, permanence, and the benchmarks that measure them.
What it costs concept · step 13
Why video is expensive and how that shapes a plan.