How generators work and which to pick

The model has seen billions of captioned pictures and learned how words hook onto images. Your prompt is a caption, and the model paints whatever fits it. Which leads straight to the one rule that matters: the picture is only as good as the description.

The 2026 map of generators

• Built into ChatGPT / Gemini — the free way in: they follow long descriptions in ordinary language and let you fix things by talking ("now drop the text in the background"). For social posts and slide decks, that's plenty.
• Midjourney (~$10/mo) — the benchmark for artistry: light, composition, an expensive-looking image straight out of the box. What designers and marketers reach for.
• DALL-E — strong at following instructions exactly and at putting readable text in a picture.
• Open models (Stable Diffusion and its descendants) — free on your own hardware, total control, steeper climb.

A rule of thumb

Start with the built-in one — nothing to pay, and you can fix things by talking. Move to Midjourney when you hit the artistic ceiling. Everything this course teaches carries over: the service changes, the language doesn't.

What generators are still bad at (2026)

• Long text inside a picture — short labels come out fine now, paragraphs don't. Lay the text on top in an editor instead.
• Accurate charts and diagrams — ask a chatbot for code or markup, not a picture.
• Specific real people — the results are bad, and the ethics and law are worse.

Try it now

Make your first picture in a built-in generator: describe a place you love the way you'd describe it to a friend — three or four sentences, with real details in them. Then look at what a three-word prompt would have given you.