Oh, So You Wanna Make Some Cute Mockups of Your House? Your Product? Your Client's New Brand? I Tested Which AI Model is the Best for Images. And, It's Not Even Clooooose.

I hate that I have to say this, but I have to say it.

Gurggley gurggley, gurggley, gurggleeeee. (That's me clearing my throat. I think.)

First, background!

I bought an old colonial farmhouse from 1872 in the middle of nowhere Pennsylvania and am currently renovating the damn thing, AKA making it look worse, probably. You should see how many gallons of paint I currently have sitting in the downstairs hallway. Walking around them is an art form. I am pretty sure that if I don't start using some of this paint, the floor is definitely going to cave in and then the staircase is going to cave in and then I am going to get sucked into the creepy old stone basement abyss with all the spiders and DIE A SLOW AND TRAGIC DEATH FROM A PUNCTURED LUNG. I'll probably land on an old rusty rake. There is a very large work table full of tools down there that, presumably, the previous owners...used??? But I refuse to touch them or even think about them, so they're just going to be there until some poor sap buys this place from me and wonders why I couldn't fucking clean anything.

ANYWAY.

So, I've decided to enlist the help of AI with some of this ~ decoratingggggg business ~, as I have the very bad habit of thinking it will look great and buying it and then realizing that IT DOES NOT LOOK GREAT.

Alas, I decided to run a little experiment and see which AI application would give me somethin' delicious.

The mission: Slap up a photo of my upstairs hallway, give it the names of the colors I'm using on the walls and doors, and then have it compare a few different wallpaper options I'm considering.

Well, whomp, whomp, whompppp, I knew I couldn't use Claude for this—my absolute favorite AI ever for running my business, specifically—because Claude doesn't have an image generator or generative image editor because Claude is a BUSINESS BADASS and Claude prefers to keep it that way. 🤣 Claude is wayyyyyy better at serious work like coding websites and doing my bookkeeping and reminding me how much I need to pay in taxes and performing business-related tasks.

However.

ChatGPT has been touted as being AWESOME at images.

And, so has Google's Gemini / Nano Banano. (Definitely wins on name.)

So, being the absolute sad sack of an AI nerd I am, I thought I'd test 'em out.

WELL.

THERE IS JUST NO CONTEST. ❌

None. Negativo. Zip. Zero.

ChatGPT's image model wins by a muddy, delirious landslide. 🛘

Here's the mockup I got out of ChatGPT almost instantly:

Correct colors.
Correct scale.
Correct wallpaper.
No weird mutations.
True to life.

Then, I gave the same prompt to Google's Gemini (using Nano Banano image technology).

This was the result ONLY AFTER having to re-prompt it approximately 7 times, because it kept trying to: a) put the wallpaper on the brown wall; put a rug on top of the bannister???; c) Use other made-up, invented wallpaper patterns (note the pattern in the back by the window seat is completely different.)

In addition:

The bannister isn't the right color...at all?
The blue doors & trim aren't the right color....at all?
The scale of the wallpaper is huge.
And, I'm most offended by the totally invented wainscotting that does not exist in that hallway at all....hahahahahahhaha.
Also, where did the doors go?

Swap "wallpaper" for "packaging" or "storefront" or "your client's new lobby" and it's the same test: can this thing edit a real photo without hallucinating a rug?

For me this is an easy decision:

While this was only one experiment, ChatGPT easily outperformed Gemini without much effort.

Next up: mocking up the rest of the house, so I can find out exactly which of the 47 gallons of paint in my downstairs hallway were a total waste of money because they'll look horrific anywhere I put them, over and over and over again, amen.

The good news: now I'll know before I open a single can.

The bad news: the cans are still sitting there. And so is the floor. For now.