Follow : Add us as a preferred source on Google.

Key takeaways

• ChatGPT Images is a major upgrade over previous image tools.
• Text rendering and recontextualization are now surprisingly accurate.
• Edits still change details but avoid uncanny valley results.

It's been less than a week since OpenAI introduced GPT-5.2. Now, it's back with a major update to ChatGPT's image generation capability. It's almost like it has an army of robots writing software for it. Oh, wait. It does.

Also: I put GPT-5.2 through a 14-round test, and the AI model raised some serious questions

This update is overdue. A week or so ago, I reached out to OpenAI about their image generation tools. I wanted to ask about how I'd noticed that DALL-E wasn't being discussed much and that the overall ChatGPT image quality didn't appear to have kept pace with Google's Nano Banana Pro.

I was told that ChatGPT's image generator was "GPT-4o image generation in ChatGPT." GPT-4o was generations behind ChatGPT's mainstream LLM, GPT-5.2. But now, just a few days after releasing GPT-5.2, OpenAI has released that overdue major update to image generation.

Continuing their trend of naming products without ever consulting a product manager, this new feature is called "the new ChatGPT Images" or, alternatively, GPT Image 1.5. We'll just let that go for now. Just know that the image generator now available in all tiers of ChatGPT, including free, is this new model.

Also: Are GPT-5.2's new powers enough to surpass Gemini 3? Try it and see

One quick note: The new ChatGPT Images feature doesn't appear to have made it to my Mac ChatGPT app yet, so for now I'm doing everything in Chrome.

(Disclosure: Ziff Davis, 's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

A first look

Today, I'm taking a first look at what the new ChatGPT Images can produce. I'll be back with a more comprehensive analysis, but let's just get a feel for what we have here.

Also: Is ChatGPT Plus still worth $20? How it compares to the Free and Pro plans

A big new feature is much more accurate recontextualization of images. That's where the AI can edit the image, composite new items into it, change the look or contents, or otherwise take a starting image and do something interesting with it.

We'll start with this image of me walking on a park path.

I uploaded that into ChatGPT and gave it this prompt: "Put the man in a red shirt with a keep calm carry on logo on the front." The result is this image. I've included the image twice. You can see the pure output generated by ChatGPT on the left. You can see my annotated version of the same output on the right.

First, of course, it transformed a wide image into a square image. I tried the assignment again with both "keep the aspect ratio" and "make it 16:9." Both were ignored. It did change the shirt. It did an absolutely great job with both the text representation and how the "keep calm" style is normally done.

But it also made some subtle and unasked-for changes. At (1), it changed my expression. My mouth had originally been in a bit of an open smile. ChatGPT gave me a bit of a grimace. At (2), you can see it changed the camera position, moving it from a somewhat overhead shot to a more straight-on shot.

Also: How to use ChatGPT: A beginner's guide to the most popular AI chatbot

At (3), it pulled the camera back, showing more of my pants leg and adding a shadow. It changed my position relative to the tree on the left (at 4), pulling me forward and pushing the tree further into the background. At (5), it added more space between my left arm and the vest and, conversely, removed space (at 6) between my right arm and the vest. That implies a subtle twist of camera location. It also removed the microphone, which was at (7) in the original image.

Still, there's no uncanny valley to this image, which is pretty impressive.

My job is too much fun

I decided to play on the idea of my T-shirt, which celebrates the classic Captain Kirk vs. Gorn fight in the original Star Trek TV series. I re-uploaded my original image and prompted ChatGPT to "put the man in front of the Vasquez Rocks." Vasquez Rocks is a famous Hollywood filming location, and it's also the setting of the James T. Kirk vs. Gorn episode celebrated by my shirt.

Also: How ChatGPT actually works (and why it's been so game-changing)

Interestingly, ChatGPT's first attempt took the previous picture it had generated and placed it in the new context. This was not what I wanted.

But when I started a new session, uploaded my original image, and gave it the same prompt, I got the correct image. What's also interesting is that this time, I got a more widescreen aspect ratio, even though I didn't ask for it. Still, it's pretty effective. ChatGPT didn't appear to change the shadows, but they do match the background image.

Then I tried, "Put a Gorn in the near foreground." And we got this. I have to say this looks very much like a composited image. The Gorn is wrong-sized. Still, it's a classic TOS (the original series) Gorn.

Also: I test AI for a living, and these 3 free tools are the ones I used most in 2025

Then, well, I kind of got carried away. I prompted, "Put the man in the yellow shirt on the right side of the image, facing left. Put the Gorn on the left side of the image, facing right. Have them face each other as if they're about to fight."

It does look a little fake (I mean, man in suit, right?). But still, the shadows are right, and I do look like I could take him. Plus, the image is so meta, what with the location and T-shirt and all. So let's take it one more step: "Put the man in a classic Star Trek original series yellow captain's uniform."

Also: How I use AI to bring my kid's art to life - and why it's a fun learning opportunity

Here, there are some issues. First, it demoted me, only giving me two stripes instead of the three stripes Captain Kirk has on his wrist as captain. Fine, I can't really argue that. But it also rolled up my right sleeve. If you look closely, you'll see it's not missing that part of the sleeve, but that it's bunched up around my elbow. That could happen in a fight.

Also: Free Photoshop in ChatGPT: How to edit photos with AI, no experience necessary

OK, this is getting to be too much fun. "Dress the Gorn like Tom Baker in Dr. Who."

Let's get even more silly. "It's winter at Vasquez Rocks and about 28 degrees F. Dust everything with snow."

Now, "Dress the man in winter gear that would be issued by Starfleet. But give him a US Navy issue knit cap."

I told ChatGPT to "decorate the area and the rocks for Christmas. Make it festive." It titled the image "Holiday showdown at Vasquez Rocks." It showed a candy cane coming out of my shoulder, but it managed to stick Santa in the distance.

Also: How well does ChatGPT know you? These simple prompts can reveal a lot

I didn't like exactly how ChatGPT was holidayifying the image, but I was amused. So I told it, "Remove the candy cane at the right of the image. Get rid of the gifts. Add a lot more lights. Move the scene into twilight so the rocks show up, but the lights also add warmth to the overall scene. Get rid of the star at the top of the rocks."

Then I asked ChatGPT to turn it into a party invitation with the prompt, "Turn this image into a holiday party invitation for a party on December 17."

Also: This new most popular AI image and video generator has enterprise users flocking to it

The first version (on the left) kind of works. Points to ChatGPT for getting all the text right. But after all the work on the image, it somehow seems generic. So I decided to use the capabilities of ChatGPT itself, giving it this prompt: "This party invite seems somehow generic. Please make it more evocative of the setting and context. Make it seem like a fun in-universe invite that fits the vibe." That's the version on the right.

And wow. It originally wrote it as "Captain Kirk vs. Gorn," but I told it to change "Captain Kirk" to "Gewirtz." And that was the result. I love it. I particularly love the headline it came up with all on its own: "Lights! Eggnog! Punching!"

A major improvement

This is a definite improvement over what ChatGPT previously offered. By far. Its text handling is excellent. Text has historically been a big problem for AI image generators.

There is a weird blurring in-and-out effect that ChatGPT shows while it's generating the image. I encourage OpenAI to remove that effect, because it can be a bit nauseating to watch. But it's fairly quick, and you don't have to stare at a blurring in-and-out image.

Also: The best AI image generators of 2025: Gemini, ChatGPT, Midjourney, and more

I'll be back with a more definitive review, but I have to say that this version of ChatGPT Images will likely live long and prosper. If nothing else, it's enormously fun.

What do you think of ChatGPT's new image generation capabilities? Have you tried recontextualizing or editing images this way? Did the results meet your expectations? How important are text accuracy and faithful edits to you? How does this compare to other image tools you've used? Comment below.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.

Artificial Intelligence

I'm a Photoshop diehard, but Canva's free tools won me over - and saved me money

AI will cause 'jobs chaos' within the next few years, says Gartner - what that means

I've tried many AI smart glasses in 2025 (including Meta Display) - these are the only ones I'd wear all day

Why AI coding tools like Cursor and Replit are doomed - and what comes next

• I'm a Photoshop diehard, but Canva's free tools won me over - and saved me money
• AI will cause 'jobs chaos' within the next few years, says Gartner - what that means
• I've tried many AI smart glasses in 2025 (including Meta Display) - these are the only ones I'd wear all day
• Why AI coding tools like Cursor and Replit are doomed - and what comes next