SSkilvy

No magic β€” just description

Image generators (Midjourney, DALLΒ·E, Stable Diffusion and others) are trained on millions of "image + caption" pairs. You give text β€” the model assembles the most fitting image.

Text promptStyle & referencesParameters Generator Image
From words to picture: what shapes the result.

Key idea: the model draws what you described in words, not what you pictured in your head. The more precise the description, the closer the result.

What to grasp right away

  • The result is probabilistic. The same prompt yields different images β€” that's normal.
  • Details decide. "A cat" vs "a ginger cat by a window in warm sunset light" β€” worlds apart.
  • Iteration is inevitable. The first image is a draft; then you refine.

Three beginner expectations

  • "It'll read my mind." No β€” describe it explicitly, or you'll get an average version.
  • "The very first image will be perfect." Usually not: budget 3–5 tries.
  • "Naming the object is enough." The object is just one of four layers (next lesson).

Exercise

Generate an image from the word "coffee", then from "a cup of espresso on a wooden table, morning light from a window, close-up". Compare β€” you'll instantly feel what details decide.

In this course you'll learn to describe an image so you get what you want in a few tries, not by luck.

🧠 Why does one prompt give different images?