Eon wanted a picture of something enormous floating over New York City. Not a photo — there's no camera that can take this shot — so he and his dad built it with AI. The plan: one AI to help write the idea down clearly, and a second AI to draw it. By the end, Eon is screaming "That's what I wanted!" at the screen. Here's how two AIs made one picture.

Step 1: turn a wild idea into clear words

First Eon had to say what he wanted, and his idea was big. Two things over a bird's-eye view of New York City at dusk: on the left, an actual-size Rolex watch as tall as the skyscrapers; on the right, a star called Stephenson 2-18, shrunk down because — as Eon put it — it's "so big you can't possibly fit in New York City. It might just eat New York City."

Instead of typing all that, they opened ChatGPT in dictate mode. Dictate means you speak and the AI types your words out for you. Eon described the scene out loud, and ChatGPT turned the messy spoken description into a tidy, detailed prompt — a written set of instructions for an image AI. The prompt it wrote back was packed: "a cinematic, ultra-detailed bird's-eye view of New York City at dusk, with skyscrapers, Central Park..." and the watch and the star in their spots.

That's the lesson hiding in this step: you don't have to remember every fancy detail yourself. One AI is great at taking a rough idea and writing it up clearly. The clearer the words, the better the next AI can do its job.

Step 2: hand the prompt to a second AI

Now the workflow gets interesting. Eon copied the whole prompt ChatGPT wrote (right-click, then scroll down to "copy"), opened a different AI image tool, and pasted it in. ChatGPT can't draw the picture — it writes words. So a second model does the drawing. Eon picked Nano Banana (Google's Gemini 2.5 image model); the tool also offered others like Firefly and Flux.

Generating cost 10 credits. Eon's dad explained that credits are basically money — they stand for computational power, the work the computer has to do to make the image. Eon tapped Generate, waited while it "thinked," and then — the picture appeared exactly as planned: Central Park in the middle, a brand-new Rolex on the left, Stephenson 2-18 glowing on the right. He clicked it to blow it up bigger. "Oh man, this watch right here, brand new!"

Two tools, two jobs: one AI for the prompt, one AI for the image.

The Sun is a star — and stars come in giant sizes

Early on, Eon made a quick "sun" and called it a planet. His dad caught it: "This is not a planet. This is a star." Our Sun is a star — stars make their own light and heat, while planets like Earth only reflect a star's light.

Then it got cooler. Eon didn't want just any star — he wanted Stephenson 2-18, which is a real star, one of the largest ever found (a red hypergiant). That's why he kept insisting it had to be "shrunk down": a star like that is so huge that, at real size, it wouldn't just sit over the city — it would swallow it. So stars aren't all the same; our Sun is ordinary-sized compared to a monster like Stephenson 2-18.

One problem, more than one tool

At the end, Eon's dad asked the best question: what did we learn today? The answer wasn't just "AI is fun." It was that there are multiple ways to solve one problem. You could build this scene by hand in Blender (real 3D modeling), or you could chain two AIs together — one to write, one to draw. Different tools, same goal. A good maker knows there's more than one path.

Try it

Make your own two-step prompt with a grown-up. First, say your idea out loud in full: pick a place (a beach? the Moon?), a thing (a giant pizza? your pet, house-sized?), a time of day, and where each thing goes — left or right. Notice how adding details makes the picture in your head sharper. Then imagine handing those exact words to a drawing AI. The clearer your describing, the closer the picture will be to what you actually wanted — and remember to use AI with a parent, because it can still make mistakes.