Making pictures with AI
How image generators turn words into new pictures, how to write a strong image prompt, and the mistakes and rules to watch for.
Type a few words, and AI paints a brand new picture. Let's see how it works. Some A I tools can make brand new pictures from a description. You type what you want to see, and a few seconds later, there it is. Many apps can do this, including ChatGPT, Gemini, Meta A I, Canva and Adobe Firefly.
How an image generator works
First, you type your words, like, a goat wearing sunglasses. The A I model learned from millions of pictures, each with a caption describing it, so it knows what goats and sunglasses look like. Then it makes a new picture from scratch. Many image generators start with random dots, and shape them, step by step, until they match your words.
New word: Image generator
A tool like this is called an image generator, because it generates, or creates, new images. The description you type is called an image prompt. And just like with chatbots, a better prompt gets a better picture.
A strong image prompt
A strong image prompt has four parts. First, the subject. Who or what is in the picture? Second, the setting. Where is it, and when? A market, a beach, at night? Third, the style. A cartoon, a photo, a watercolour painting, or 3D animation? And fourth, the details, like colours, the mood, or anything else you want to include. Here's a strong image prompt. A girl in a yellow dress, flying a kite on a Lagos beach at sunset, in a colourful cartoon style. It has the subject, the setting, the style and the details. A few seconds later, you get a new picture. If it isn't quite right, change your prompt, and try again.
Watch out for
Image generators still make mistakes. Watch out for odd hands, with extra or missing fingers, strange writing, where words in the picture come out jumbled, and wrong numbers. Ask for three goats, and you might get four. Two rules. First, be honest. If a picture was made with A I, say so, especially if it looks like a real photo. And second, never make pictures of real people without their permission. Fake pictures can hurt people, and you'll learn more about that in Unit 6.
Quick check
Quick check. Which image prompt is the strongest? The first one. It has a subject, a goat in sunglasses, a setting, a busy Lagos market, and a style, bright 3D cartoon. Options: A goat wearing sunglasses in a busy Lagos market, bright 3D cartoon style; goat; A picture; Make something nice.
Recap
Image generators make new pictures from words. A strong image prompt has a subject, a setting, a style and details. Check for odd hands, strange writing and wrong numbers. And label A I pictures, and leave out real people.
Key points
- Image generators create new pictures from a written description, called an image prompt. Many apps can do this, including ChatGPT, Gemini, Meta AI, Canva and Adobe Firefly.
- They learned from millions of pictures with captions, and many build each new picture step by step from random dots.
- A strong image prompt has a subject, a setting, a style and details. Example: a girl in a yellow dress flying a kite on a Lagos beach at sunset, colourful cartoon style.
- Watch out for mistakes like odd hands, jumbled writing and wrong numbers of things.
- Label AI pictures, and never make pictures of real people without their permission.
Try it: Build an image prompt
Sort each phrase into the part of an image prompt it belongs to.