Gemini’s Nano Banana trimmed my beard and gave me a sunburn, but I still love it

Google released its Nano Banana image editor a few weeks ago, and I’ve been playing with it ever since. It fixes the biggest issue I had with image editing in Gemini: whenever I tried to make a small adjustment with a prompt, the AI would generate a completely new image.

Not anymore. With Nano Banana, I’ve successfully changed tons of small details on various images, all while leaving the rest of the elements unchanged. It works, but not always. As usual, there are a few quirks, all of which I’ll discuss in detail in this post.

I recently went on a weekend trip to Lake Bled here in Slovenia and, as usual, took way too many photos. I figured they would be perfect for testing how good (or bad) the new image editing model really is.

I uploaded a picture of myself standing next to the lake, with the castle in the background, and got to work. I used a separate prompt for every single change I wanted to make instead of trying to make multiple changes at once.

I asked Gemini to replace my smartwatch with a Rolex Datejust, turn my shirt navy blue and add a Louis Vuitton logo to it, switch out my glasses for more modern white ones, and give me a few tribal tattoos for a more rugged look. I even asked it to make the water bluer and place a boat with two people on the lake.

Overall, Nano Banana did an excellent job. I’m impressed, especially when compared to the previous model. It never changed anything else in the photo, except for one occasion, which I’ll get to in a second. It works fast, and most of the initial changes were so good that no one could tell it was AI at work. However, its limitations quickly started to show up.

I ran into all sorts of issues during this test. For example, the AI got confused and placed the tattoo on my right arm instead of the left one as I asked. It took me quite a few tries with more detailed prompts to get this right, although I eventually decided to get them on both hands. It also had issues when I prompted it to remove the tattoo; it didn’t do anything at first, but it handled the request after a few more prompts.

For whatever reason, Nano Banana also refused to make me more muscular, despite multiple attempts. It’s hard for me to gain weight and put on muscle, and it looks like not even AI can help. I’m a lost cause.

One of the biggest issues I had was when I asked the AI to make the forest around the lake greener. While it technically did it as you can see in the image above, it also removed the castle and the island with the church on it — yes, that’s actually an island — and placed a big mountain in the background. What’s funny is that it’s the same mountain range I’m looking at from where I’m standing in the picture, so it seems Gemini grabbed it from one of my other images in Google Photos and placed it in this picture. I wasn’t a fan, so I asked it to revert the changes, which it did without issue.

Other issues include the castle on the hill, which the AI turned into something that resembles a shack. Then there’s my beard, which got trimmed for no apparent reason. And, just for fun, Gemini also decided to give me a sunburn on my face. Sure, it was a very hot day, but not that hot.

The last thing worth mentioning is the image quality. The more changes the AI made, the grainier the image became, which is evident when you compare the first and last pictures.

The grainy result from the first test threw me off, and I wanted to figure out what caused it. I suspected it was because I requested too many changes, especially on just one part of the image — myself.

With that in mind, I uploaded the same image again and requested all the changes in a single, comprehensive prompt to see what difference it would make. And it made a big one. The resulting image was less grainy, although the problem was that not all the changes were made.

My watch wasn’t replaced with a Rolex, my shirt’s color didn’t change, and the water wasn’t any bluer. But hey, at least Gemini figured out which arm is the left one this time, so kudos for that.

The glasses also came out blue instead of white, but that was my mistake in the prompt. I then gave the AI a few individual requests to change the color of the glasses, make the water and my shirt bluer, and put that gorgeous Rolex on my wrist. Nano Banana was able to get everything done but refused to give me the watch I wanted, even after several more prompts.

However, the important part is that while the final image was still a bit grainy, it was less pronounced than in the first test. I then took it a step further and gave myself a chain around my neck and a new haircut. I also prompted the AI for a thicker beard, mainly because it trimmed mine during the first test. I may have overdone it, but I had a blast with the tool and am generally happy with the result, despite it not being perfect.

I wanted to shift the focus off of me and try making changes to a nature shot. I went up to the castle on the hill, took a picture of the lake, and got to work.

Since using fewer prompts seemed to produce a less grainy picture, I requested several changes from the get-go. Those included bluer water, an eagle on the ledge, a few boats on the water, and mountains in the background. I was especially surprised by how seamlessly the AI placed the mountains behind the hills, making it very hard to tell it wasn’t a real photo. The boats seemed a bit generic, so I asked it to make them more varied, which it did. I wasn’t completely happy with the result, but it looked better.

I also asked for a paraglider in the background and a waterfall on the hill behind the island church. There were no issues with the paraglider, but the waterfall was placed on the wrong side and looked a bit off at first, with water dropping straight down from the trees. I asked Gemini to make it more realistic by adding a rock wall behind the waterfall, which it did. A detailed eye can easily spot that it’s an AI-generated image, but it’s not a bad attempt.

I could have played around with it more, but I was generally happy with the end result. There’s definitely room for improvement, though.

One of Nano Banana’s highlights is its ability to combine two images into one. I tried this multiple times, and I think Gemini didn’t do a terrific job in this area.

I uploaded an image of myself and my partner and asked the AI to create a picture of us walking beside the lake, holding hands. As you can see, we’re not holding hands. While we are by the lake, the AI decided to keep the first image intact and just paste me into it. I look out of place, and the image doesn’t look natural to my eye.

I then prompted it to make an image of the two of us having dinner together. This one came out the best, even though we don’t look exactly like ourselves. A good try, though.

Unfortunately, it went downhill from there. The next two images were pretty bad, as we don’t look anything like ourselves. In one, we’re running a marathon, and in the other, we’re dancing at a party. I was not impressed since there’s no resemblance to us.

Nano Banana is supposed to be able to change the scene of a picture completely. For example, it can take a photo of you and then turn you into a lawyer sitting in a courtroom while keeping your face intact. I tried it and hoped to get better results than with the previous test, but Gemini fell short again.

I uploaded a picture of myself and asked to be turned into a pizzaiolo, an NBA player, and a matador in an arena. While all the images look great, the person in the photos doesn’t look anything like me. Well, perhaps there are a few resemblances, but they’re generally way off. The NBA one was the best, while the pizzaiolo came out quite bad. I don’t look anything like that, but hey, at least I got a fancy watch this time without even asking for it.

I experimented with close-up images of my face so the AI could have a better chance of nailing my look, but I had no success.

Just so you know, the editing capabilities aren’t limited to the images you upload. You can also ask Gemini to create an image from scratch and then make small changes to it without altering the rest of the scene.

As an example, I created an image of a cat walking down a forest path and then changed its color to white and the tip of its tail — and paws — to black. I also changed the color of its eyes and switched to a winter theme. This all worked perfectly, as you can see from the images above.

Nano Banana has its faults. I don’t like the grainy pictures it produces at times or the fact that it trimmed my beard and gave me a sunburn for no apparent reason. Based on my tests, I’m not a fan of its ability to combine two images in most cases, and changing the scene of a photo with a person in it didn’t yield great results.

But to be honest, I expected many of these issues. This is Google’s first real attempt at in-image editing without regenerating the entire scene, and I am impressed with it overall. Based on how fast AI tools are improving, it’s going to get even better very soon.

Generally speaking, it’s a very useful tool that can quickly and efficiently make small changes to your images. I have noticed that the more detailed a prompt is, the better the results. The good news is that if you don’t like a change that was made, you can easily revert it and try again with a new prompt.

The overall success seems to depend on a combination of the image itself, the requested changes, and the prompts used. You have to play around with it to get it right, but when you do, the results can be amazing. Just imagine how long it would take you to change a shirt’s color or replace a pair of glasses with traditional editing methods. Even then, the results might be worse than what Nano Banana can produce, depending on your skill level.

So yes, I am generally impressed with Nano Banana, and I definitely recommend you give it a try — it’s free. Most people won’t request dozens of changes to a single image but will likely stop at just a few. In that scenario, the image quality doesn’t deteriorate as much, and the final result doesn’t scream “AI.”