OpenAI Just Dropped GPT Image 1.5 and It Is Honestly Kind of a Trip
OpenAI just updated ChatGPT's image generation and we put it to the test with some Star Trek chaos and holiday party invites.
- neuralshyam
- 5 min read
Is it just me, or does it feel like OpenAI is currently operating on about fifteen pots of coffee and no sleep? Seriously, it’s been about five minutes since GPT-5.2 landed, and they’ve already circled back to drop a massive bomb on the image generation side of things.
For a while there, ChatGPT’s image game was feeling a little… dusty. While Google was out there doing wild stuff with their Nano Banana Pro (real name, apparently), ChatGPT was still rocking older tech that felt a few steps behind the main brain of the LLM. But the “new ChatGPT Images”—or GPT Image 1.5, if you want to be technical—is finally here to fix that.
I spent the afternoon messing around with it, and honestly? It’s pretty impressive, slightly weird, and 100% more fun than the old version.
The “Everything Has Changed” Update
The biggest thing you need to know is that this isn’t just about making “prettier” pictures. We’ve had pretty pictures for a while. The real flex here is recontextualization.
In plain English: you can give it a photo of yourself, or your dog, or your messy kitchen, and tell it to change specific things without the whole image melting into a puddle of AI artifacts. It’s basically like having a professional Photoshop editor living inside your chat bar, except this one doesn’t charge you $50 an hour and actually listens to your dumbest ideas.
The best part? It’s rolling out to everyone. Free users, Plus users, Pro users—everyone gets a taste of the new pixels.
Putting the “Edit” in Editing
To see if this thing actually works, I started with a basic shot: a guy walking down a park path in a yellow shirt. Simple, right? I asked it to swap the shirt for a red one with a specific “Keep Calm” logo.
The results? The text was perfect. If you’ve been following AI for more than a week, you know that AI usually treats the English alphabet like a Suggestion Box written in ancient runes. But here, the “Keep Calm” text was crisp and correctly styled.
However, AI still has its “main character” moments. Even though I asked for a 16:9 widescreen shot, it gave me a square. It also decided I needed a slightly different facial expression—turning a smile into more of a “I just realized I left the stove on” grimace. It moved the camera angle, adjusted the shadows, and even deleted a microphone that was in the original shot.
Is it perfect? No. Is it way better than what we had? Absolutely. There’s no “uncanny valley” creepiness where the skin looks like wet plastic. It just looks like… a guy in a red shirt.
Let’s Make It Weird (The Star Trek Test)
Since the AI was playing nice, I decided to take things to the absolute limit. I’m a bit of a nerd (shocking, I know), so I decided to transport myself into the world of Star Trek.
I took that park photo and told ChatGPT to put me at Vasquez Rocks—the famous spot where Captain Kirk fought the Gorn (that green lizard guy in the rubber suit).
- Phase One: It nailed the background. The lighting on my shirt actually matched the desert sun.
- Phase Two: I told it to add the Gorn. It looked a little like a collage at first, but it got the “classic TV” vibe exactly right.
- Phase Three: I asked for a full Starfleet uniform. It gave me the gold captain’s shirt, though it demoted me to a lieutenant by only giving me two stripes on the sleeve. Rude, but I’ll take it.
By the time I was done, I had the AI adding snow to the desert, dressing me in winter Starfleet gear (complete with a US Navy knit cap), and then—because why not—decorating the whole alien battlefield for Christmas.
“Lights! Eggnog! Punching!”
The real “aha!” moment came when I asked it to turn the final chaotic image into a holiday party invitation.
Normally, AI-generated invites are a mess. They usually say something like “HAPY HOLIDY AT THE VORP.” But GPT Image 1.5 handled the text like a pro. When I told it the invite felt a bit “generic,” it actually got creative. It changed the headline to: “Lights! Eggnog! Punching!”
It’s that kind of personality that makes this feel less like a tool and more like a collaborator. It understood the “vibe” of a guy fighting a lizard man in a Santa hat and leaned into the joke.
The Reality Check
Look, it’s not all sunshine and perfect pixels. There’s this weird “blurring” effect that happens while the image is generating. It’s supposed to look cool, I guess, but it mostly just makes me feel like I need a new pair of glasses. I’m hoping they ditch that in the next patch.
Also, it still struggles with following specific technical instructions like aspect ratios or keeping every single detail of your face exactly the same. You might go in looking like yourself and come out looking like your slightly more handsome cousin.
Is It Worth Your Time?
If you’re just using ChatGPT to write emails to your boss, you’re missing out. The new image capabilities are a legitimate playground. Whether you’re trying to visualize a scene for a story, making dumb memes for the group chat, or actually trying to design an invite, the “recontextualization” feature is a game-changer.
It’s fast, it’s smart, and it actually understands what a Christmas tree looks like in the middle of a sci-fi desert.
So, go ahead. Upload a selfie and tell it to put you on Mars wearing a tuxedo made of cheese. At the very least, you’ll get a laugh out of it.
What are you guys planning to make first? Are you more into the realistic edits or the “lizard-man-at-a-Christmas-party” level of chaos? Let me know—I’m curious to see if anyone can break this thing.