Student Stories

How I Built My First Poster with AI (and What I Changed)

A student walks through their STEMFest project, what the AI got right, and the parts only they could add.

A student walks through their STEMFest project, what the AI got right, and the parts only they could add.

No headings found

HOW I BUILT MY FIRST POSTER WITH AI (AND WHAT I CHANGED)

I needed a poster for STEMFest by Friday and I'd never made one before. So I did what everyone does now: I opened an AI image tool and typed "poster for a science fair." What came back looked fine for about three seconds, then I noticed everything wrong with it.

THE FIRST DRAFT

The AI gave me a poster with a generic beaker graphic, a made-up event name, and text in a font I couldn't actually read past ten feet. It wasn't bad, exactly. It just wasn't mine. It was the average of every science fair poster the model had ever seen, which is a real problem when you're trying to stand out at an actual event.

That's when it clicked: I hadn't told it anything specific. No event name, no date, no colors, no vibe. I'd typed a search query, not a set of instructions.

WHAT THE AI GOT RIGHT

Once I rewrote the prompt with our real event name, our school colors, and the date, the second draft was genuinely useful. The layout hierarchy was solid, the AI knew where a title should sit versus a subhead, and it handled the typography pairing better than I would have on my own. It also gave me three variations in the time it would've taken me to open Illustrator.

For the boring-but-important stuff, spacing, contrast, balance, it was a faster starting point than a blank canvas.

WHAT I HAD TO CHANGE

The details were where it fell apart. The QR code it "generated" didn't link anywhere. The team names were spelled wrong. A robot graphic had six fingers on one hand, which nobody on my team noticed until we'd already printed a proof. I had to manually fix every piece of actual information on the poster, because the AI doesn't know facts, it predicts what plausible text looks like in that spot.

I also toned down the color scheme. The AI defaults toward loud and busy; I wanted something a judge could read in five seconds walking past, so I cut two colors and doubled the white space.

THE PARTS ONLY I COULD ADD

The thing that actually made the poster ours wasn't anything the AI did. It was the one line at the bottom explaining what our project actually solved, written in our own words, based on three weeks of testing that no model could know about. That sentence is the whole reason the poster existed. Everything else was scaffolding.

WHAT I'D DO DIFFERENTLY

Next time I'd start with the real details on day one instead of a vague prompt, and I'd treat the first draft as a rough layout, not a finished poster. AI got me to "close" fast. Getting from close to correct was still on me.

Ready to think like a reasoning engineer?

2Skills helps young people learn how to identify meaningful problems, choose the right AI tools and prompt them clearly to build real solutions.

Task list of learning how to use AI tools

Ready to think like a reasoning engineer?

2Skills helps young people learn how to identify meaningful problems, choose the right AI tools and prompt them clearly to build real solutions.

Task list of learning how to use AI tools

Ready to think like a reasoning engineer?

2Skills helps young people learn how to identify meaningful problems, choose the right AI tools and prompt them clearly to build real solutions.

Task list of learning how to use AI tools

Ready to think like a reasoning engineer?

2Skills helps young people learn how to identify meaningful problems, choose the right AI tools and prompt them clearly to build real solutions.