How do text-to-image AI models generate unique artwork?

Text-to-image AI models are like colorful recipe books that create artwork from words, just like you make a cake from a list of ingredients.

Imagine you have a robot chef who knows how to mix colors and shapes. When you tell it, "Draw a yellow duck wearing sunglasses," the robot chef looks up the recipe for ducks and sunglasses in its book. It mixes the right colors, adds some shapes, and boom! You get a picture of a happy yellow duck with cool sunglasses.

The robot chef learns by watching lots of pictures and reading what they’re called. For example, it sees many ducks, so it knows how to draw one. It also learns that "sunglasses" mean two small black shapes on top of eyes.

When you give the AI a new idea, like "a blue cat flying through space," it uses what it has learned to make something completely new, not just copying from old pictures, but creating a fun and original picture that matches your words.