I’ve been keeping an eye on it. AI Anyone who works with image editing models should have heard of it. Nano BananaRecently, it’s been an incredibly hot topic in the AI circle. No one currently knows who developed Nano Banana, but some speculate it’s Google There’s a newly developed model that was secretly in the works, and I’ve been playing around with it these past few days. I have to say, it’s absolutely insane — it completely crushes other AI editing tools. Not only can it edit images, but it can also merge multiple images into one, and even change styles. The results it produces are fantastic, and it’s totally free. You really have to try it.
Nano Banana is currently the most insane free AI image editing tool—retouching photos, merging images, and changing styles are all incredibly powerful.
LMArena is a platform created by researchers at the University of California, Berkeley, designed to let everyone freely compare, interact with, and evaluate various AI models. After you submit a prompt, the system automatically pairs models and shows responses from two anonymous models. You then choose the better answer, and only after you’ve made your selection does it reveal which specific models they were. Nano Banana is one of the models hidden in image generation. It is completely free to use and requires no account registration.
After entering LMArena via the link above, click the image icon below:
This will enable the image generation feature, at which point you can enter prompts and upload reference images:
You can enter prompts in Chinese, but AI models still understand English better, so if possible, English input is recommended. Let me first test simple text-to-image generation.
Next, two results will appear for us to rate. Before rating, only Assistant A and Assistant B will be shown above the image; we won’t be told what models they are. That will only be revealed after rating. You can choose Left is Better, It’s a tie, Both are bad, and Right is Better:
I’ll go with the right side. This one was generated by nano-banana, and the result is clearly better than gemini-2.0-flash on the left. If you like it, click the download icon in the top right corner to save it:
Next, I’ll test adding other things based on this image.
One thing to note is that LMArena can’t continue answering based on previous context, so you’ll need to feed it the content again.
I asked to add a Tyrannosaurus rex behind the car, lowering its head and approaching the man in front:
The two images generated the first time were both just so-so, so even before any ratings came in, I already guessed neither one was a Nano Banana:
As expected, after scoring, neither of them was right. The left one is gemini-2.0-flash, and the right one is flux-1. At this point, all I can do is request another generation and see if the next one will match the Nano Banana model.
The image on the right turned out great on the second generation, and the colors looked natural too, so I’m guessing it’s Nano Banana:
After scoring, it really was indeed:
On the left is the original image, and on the right is the T-rex added using Nano Banana. The color contrast, the realism of the T-rex, and its pose are all excellent, and the details of the original image are preserved with almost no changes.
Besides adding new things to an image, you can also combine two images. For example, below we combined the 電腦王阿達 figurine image with a Google phone, and asked it to generate a street view image with Taipei 101 in the background. Nano Banana generated it perfectly.
Next, I also tested putting on clothes. Although it can’t be said to be perfect, the proportions are a bit odd, and the colors are too, but indeed I really changed into a suit, still keeping the crossed-hands pose (suit image source:Amazon Japan）：
The image below is also created by combining two images with Nano Banana. The key point of the prompt is to create a professional product photography image, and the result is almost perfect. This is very useful for those who need to create product showcase images:
Replacing objects in the original photo is also possible; I request that the chocolate cake in the middle be changed to the mug I provided:
The image below was created by combining a portrait photo of Ada with a figurine image, with the scene set at “playing with the figurine at a computer desk, while a horror movie plays on the screen.” Although Ada’s face looks like a different person, everything else was generated perfectly, especially the figurine.
Next, I also tested a different approach, asking Nano Banana to generate four different photo poses. Didn’t expect that to work too:
Finally, it’s generating a poster in the style of “Just Do It,” and this one looks awesome:
Nano Banana can do even more AI image editing than this—you should take some time to play around with it.
Source: KOCPC Chinese