At least as far as I’m concerned, ChatGPT has been ruling the roost for image generation in AI ever since its native image generation capabilities were added in March this year, sparking all sorts of viral image crazes, like the Studio Ghibili one, for example.
Since then Google has taken a bit of a step back, but has been slowly working on its own rival, called Nano Banana. Last week it added its new image editor to Gemini and made it available to everybody, and now I think it has become my favourite image editor, even over ChatGPT. Here’s why.
One banana, two banana
Apparently the name Nano Banana was the anonymous codename for Google’s Gemini 2.5 Flash Image model during its testing phase on the LMArena leaderboard, and over time it stuck.
You can access Nano Banana in three ways, either through the Gemini app, selecting ‘Create images’ from the drop-down Tools menu in the prompt bar (you’ll know it’s now using Nano Banana because there’s a new banana icon next to ‘Create images’). Or you can go to Google’s AI Studio and use it there, or you can head to its own dedicated website.
I’ve been playing with Nanao Banana all weekend and can report that it’s very good in three main areas: Character consistency, realism and image-to-image fusion.
Character consistency
First of all, let’s look at character consistency. If I ask Gemini 2.5 Flash to “Draw a cat in a historically accurate Roman Centurion helmet“ it produces this magnificent-looking kitty:
It’s also a nice bright, clear image. If I then say: “Now make that cat appear in the coliseum”. It does it, and the cat and the helmet look the same:
With ChatGPT, I say: “Draw a cat in a historically accurate Roman Centurion helmet” and it produces this:
It’s got that characteristic, darker feel that ChatGPT seems to default to, but it is indeed a cat in a Roman helmet, albeit a wonky one.
When I ask it to, “Now make that cat appear in the coliseum”, it produces this:
If you look closely you’ll see the helmet is different now. It’s little consistency details like this that Nano Banana seems better at.
Realism
If I upload a picture of myself (I used the one down below in my profile), then ask Gemini Flash 2.5 to put me on a mountain, I get a picture that actually looks like me, on a mountain:
Whereas ChatGPT makes an image that looks like an AI approximation of me (it doesn’t help that it’s clearly added a few pounds!):
Not only am I fatter, but the image looks less real.
Combining images
Nano Banana can combine images together in a realistic way. For example, I uploaded the image of myself and some fence panels I put up at the weekend and asked Gemini to combine them together, while keeping the background image the same. In Gemini’s version, the background picture hasn’t changed at all, it’s just added me into it:
In contrast, even though I asked ChatGPT to keep the background the same, it reproduced a version of the background that was similar to the original, but with differences I could pick out.
Not to mention that its version of me looked obviously made by AI. You could say that the ChatGPT image had better framing as a picture, but it didn’t look as realistic.
One clear winner
For me Gemini is now more useful than ChatGPT for creating images that look real. And I think that’s what most people want AI images to be. All this is without mentioning one of the best things about Nano Banana in Gemini, which it that it’s fast.
It normally takes Gemini up to ten seconds to generate an image, whereas it can take ChatGPT up to a minute to make an image. That’s a lot of waiting around.
Of course, dedicated image creation tools like Midjourney will still be king of AI image generation for professionals, but you need to pay to use those. If want something that’s very quick and gets the job done in the most realistic way possible, then for me, Gemini’s Nano Banana is the clear winner.
