I was eager to try out the new ChatGPT 5 to see if it could use emojis to produce the kind of scene that I want for a game I’m working on. Here’s a look at how my iterative process works.
Here’s the Prompt
This is a more challenging task than it seems at first glance, because the quality of the scene I'm looking for is high enough to use in a wholesome kids game.
Use HTML/CSS/JS to create a square canvas area for a phone game. It is a tableau of a large emoji of a panda as the central protagonist within a jungle of bamboo and other tree emojis, grass and bushes of various sizing to make a beautiful scene. The panda is very large, occupying about 1/3 of the total area of the canvas.
The Follow Up Prompt
Rate the quality of the game scene, and then decide if you want to change anything to make it 10/10. For example, a Parallax effect could give better perspective than randomly placed trees, particularly ones that are out in the sky.
And from there I did some custom feedback depending on how the model did.
ChatGPT
Then this
At this point I gave specific, custom feedback and instructions.
Things I like about this scene:
* The Panda is very large
* A gradient looks great, with a clear distinction between sky and ground.
* The parallax works much better than the scattered approach from earlier.Things that need work.
* The panda is largely hidden behind the ground.
* the trees and other items in the front should be large, whereas in the distance the objects should be far
* The responsive parallax effect, while impressive, isn't actually needed. What is important is perspective for a static scene, rather than dynamically animating and repositioning objects.
Make close up trees much larger so that they reach up into the sky, put a sky in the sun, and great job keeping everything behind the panda, maintain him as the central component, but about 2x as large as the current display is.
Claude’s Code
Then this
Focusing on the most important part (the Panda is pulsing by the way)
So now it’s time for the plot twist!
Neither model perfectly nailed my vision, but I think Claude’s came closest. So I did what I often did. I took Claude’s raw code and pasted it into Gemini 2.5 (which just utterly failed my initial prompt twice), and asked it to improve the scene.
I feel like it really did, because there’s now a butterfly buzzing around in front of the panda, and a bird that flies across the sky! It’s got enough of the wholesome fun vibe that I feel pretty good about where it ended up.
Instead of pasting raw code into my stories, from now on I’ll just share the link as a Codepen. Check it out and let me know in the comments which one you liked!
Jesse, thanks for sharing your experiments! I agree, sometimes you need to superimpose AI outcomes to get the result you're aiming for!