OpenAI hyped the launch of ChatGPT Images 2.0 like it was more than just another step up in AI image generation. It treated it like a complete shift in what these tools can do. But is it actually better than Gemini's Nano Banana 2? Let's find out.
Rather than testing either tool on abstract prompts or fantasy scenes, I decided to see which was best by focussing on something more practical. The images created from scratch by AI are impressive, there's a lot more interest in what the models can do with actual photos by improving them in ways that still feel real.
I'm not talking about just adding an Instagram filter, but actually radically altering the image, changing the lighting, adjusting the mood, or even shifting the season. These are the kinds of edits people actually want, and they’re also where things tend to fall apart for AI — faces start to look off, lighting stops making sense, and scenes feel stitched together instead of whole.
To use ChatGPT Images 2.0 you simply ask ChatGPT to create an image. It's the same with Nano Banana — just ask Gemini to create an image. So, to test them I ran both AI tools through the same set of edits, using the same starting images. Here's how I got on.
Lighting test
I started with a basic studio portrait on the left, then asked both models to "replace the background with a park at the golden hour." ChatGPT’s version in the middle feels slightly unreal, though the light and shadow do make it look like the chair was actually taken outside. You can see the sunlight hitting from one direction, casting shadows across the path that match the trees and my face. And the chair does sit on the ground with proper weight, matching the perspective and texture of the path.
Nano Banana 2’s version on the right leans much harder into the “golden hour” look. The colors are deeper and more orange, and it’s definitely eye-catching. But it starts to feel more like a stylized edit than a real scene. The lighting on my face is a bit too even compared to the strong sunlight in the background. The face is smoothed out, too, to add a glow, which looks nice but makes it feel less like a real photo taken at that moment.
Winner: ChatGPT
Cinematic style test
Next I wanted to see how the two models interpreted more of a mood than a direct environment request, so I asked them to make the portrait "look more cinematic and dramatic."
ChatGPT’s version on the left feels like a real studio setup that’s been pushed just a bit further. The warm backdrop has that soft, textured look you’d get from an actual photography background, and the shadow pattern across it feels like light coming through a window or a flag. Nothing feels overdone, which is what makes it work.
Nano Banana 2’s version on the right went in a different direction altogether. It changed the photo to an entirely fictional camera facing right on my face, with a tightened frame and darkened environment. It leant into a much moodier, almost film noir style. It set up a dim interior with books and wood paneling, which is a cool idea, but it is also less real, and it has a little more of the uncanny valley look to it.
Winner: ChatGPT
Product reality test
I then moved to a basic photo of a pair of AirPods, and asked the two models to give them a marketing twist. I wrote, "Use this photo of AirPods and make them look really nice while setting them in a real-world backdrop."
Both went for a desk with coffee and notebooks, but from there the differences emerged.
ChatGPT’s version on the left feels like a real desk someone actually uses. The lighting is a little uneven in a good way, with sunlight coming through a window and hitting different parts of the surface at slightly different intensities. The reflections on the AirPods are subtle and actually react to the objects nearby, especially the darker phone and the plant. Even the wood grain has variation and small imperfections that make it feel like a real surface rather than a digital one. Plus, it even went for synergy by putting an iPhone next to the device.
Gemini's Nano Banana 2 version on the right looks more like a polished product shoot. Everything is neatly arranged, the notebook and glasses are placed just so, and the lighting is very clean and balanced. There are also random earphone cases in the background that don't look like AirPod cases, not an ideal look for marketing a specific product.
The AirPods themselves are another giveaway. In ChatGPT’s image, they pick up tiny environmental cues, with reflections and shading. Nano Banana 2 again came up with a fictional camera to make a head-on image that edges into unreality.
Winner: ChatGPT
Seasonal change test
This one is a great example of how both tools handle environmental changes, not just color tweaks. I used the photo of a lawn in full greenery and asked the two to "change the season to autumn."
ChatGPT’s version in the middle doesn’t just change color; it becomes uneven, with patches of green still peeking through and leaves scattered in a way that feels random. Some areas are more covered than others, like the wind has pushed leaves across the yard. The trees shift naturally into fall tones.
Nano Banana 2’s version on the right goes for a cleaner, more uniform transformation. The leaves are spread much more evenly across the lawn, but not all the trees have changed color. The lawn is also greener, closer to its springtime color.
Winner: A draw
And the winner is...
What stood out from all the tests was that Nano Banana 2 was usually faster, but ChatGPT Images 2.0 kept edging ahead. Even when a blind test was conducted with a couple of friends who were told only one was digitally altered, they all consistently picked ChatGPT as the one they thought was real. Though it takes a little more time, the model does seem to pay closer attention to how light behaves, how textures interact, and how faces actually look in different conditions.
The difference isn’t always obvious at first glance, and Nano Banana 2 is really more than good enough in most circumstances, but it starts to become more obvious the longer you spend with the images.
Nano Banana 2 makes images that look great, but ChatGPT Images 2.0 makes images that feel more real.
Overall winner: ChatGPT
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
