lesson
35 min

How Image Models Work

A plain-language look at how text becomes a picture, and how to write image prompts with intent.

What you will learn

By the end of this lesson, you can explain how a diffusion model makes an image and write an image prompt with clear parts.

From static to picture

Most image tools today use a diffusion model. A diffusion model starts with random static, like a TV with no signal. Then it removes the static step by step until an image appears.

Your prompt guides each step. The model turns your words into numbers that stand for meaning. Those numbers push the cleanup toward "a pan of tamales" and away from "a pile of rocks."

You do not need the math. You need three ideas:

  • The model learned from huge sets of online images. It repeats what was common there, including stereotypes.
  • It does not know facts. It knows what things tend to look like. Signs, logos, and text often come out wrong.
  • Each run starts from different static. The same prompt gives a different image each time, unless you lock the starting point. You will learn about that in Module 5.

Google Whisk, Google Flow, and Stable Diffusion all use some version of this idea.

The parts of an image prompt

Text prompts used five elements. Image prompts use visual parts:

PartWhat it controlsExample
SubjectWhat is in the pictureA clay pot full of corn husk tamales
SettingWhere it isA tiled kitchen counter, morning
CompositionFraming and angleOverhead view, pot centered, empty space on the left for text
LightMood through lightSoft window light from the right
StyleThe lookFlat illustration, warm earth colors, simple shapes
ExclusionsWhat to leave outNo text, no people, no logos

Order matters. Most tools pay more attention to words near the start. Put your subject first.

Write with intent

Weak image prompts pile on words like "beautiful, stunning, 8k, masterpiece." These add little. Describe what you actually want to see.

Weak:

Beautiful amazing Fiesta background, ultra detailed, best quality.

Strong:

Flat illustrated border of papel picado banners in pink, turquoise, and yellow, hanging across the top third of the frame. Plain cream background below for text. Simple shapes, no gradients. No people, no words.

The strong prompt says what is in it, where, and what to leave out. It also fits the brief: a border, not a fake photo.

The judgment check

Look at every image before you use it:

  • Are there garbled letters or fake logos? Remove them or regenerate.
  • Does it show a stereotype? Change the prompt, or drop the image.
  • Does it look like a real place or product it is not? Label it, or do not use it.
  • Would the client be proud to post it?

Image prompt builder

Write an image prompt for one AI-allowed frame in your storyboard. Use the Role box for Subject and Setting, the Task box for Composition, Context for Light, Format for Style, and Tone for Exclusions.

Who should the AI act as?

What exactly should it do?

What does it need to know?

What shape should the answer take?

How should it sound?

Your prompt

Fill in the fields to build your prompt.

Do not paste private details (full name, address, ID numbers) into any AI tool.

Try this: Paste your prompt into Google Whisk or another image tool. Generate four images. Write in your Prompt Library which one fits the brief best, and why.

Go deeper: Stable Diffusion, Simply explains the parts of a diffusion model in more detail.