Lesson 1 — How Image Models Actually Think
The model is not a search engine and not a designer. It is a probability machine, and it reads your first words loudest.
- Write every prompt on the seven-slot spine: Subject, Action, Setting, Lighting, Composition, Camera/Lens, Style.
- Replace subjective adjectives with nouns a camera can see: swap 'premium feel' for 'wet black slate and hard rim light'.
- Never rely on negation. Describe the state you want, not the thing you are trying to exclude.
- Actionable now: take your last vague prompt, rewrite it on the spine with an explicit aspect ratio, and run both side by side.
A diffusion model starts with pure static and removes noise step by step until a picture appears. It is not pasting stock photos together. It is predicting what an image matching your words would most plausibly look like.
Attention is heaviest at the front of your prompt. Whatever you name first becomes the subject the model protects. So lead with the product or the person, not with the mood, not with the vibe you hope to land.
Cool, premium and futuristic mean almost nothing to a model. They average a million images. Brushed aluminium, wet black slate, hard rim light are instructions. Trade every adjective you can for a noun you could photograph.
Saying no hat still pushes hats into the image. Most models have no reliable not operator inside the prompt. Instead of listing what you hate, describe the thing that fills that space: bare head, clear desk, empty wall.
If you never state an aspect ratio, most tools quietly fall back to square, and your vertical placement gets cropped into garbage. Say sixteen by nine, nine by sixteen, one by one, or four by three, every single time.
Your prompt spine has seven slots. Subject, action, setting, lighting, composition, camera and lens, then style. Write them in that order. Any slot you leave blank, the model fills for you, and it picks the boring option.
Loading the quiz…