Did you order Banana? Let's check the capabilities of Gemini 2.5 Flash Image Nano Banana

On last week, Google introduced an updated image generator Gemini 2.5 Flash Nano Banana. The web application allows step-by-step conversion of user`s images, thus replacing Adobe Photoshop for non-professional users. Today, we will test its capabilities in various scenarios with sample queries for readers.

Remove text, cut an object, and merge images

Let’s take two posters as an example: S.T.A.L.K.E.R. 2 and the new Metal Gear Solid Delta: Snake Eater. The idea is to show the gradual process of merging the two images and learn how to use Nano Banana on a basic level.

First, let’s look at the captions on both posters. Download the poster (+ sign) from MGS Delta and ask to remove the inscriptions. In the next message, download S.T.A.L.K.E.R. 2 and repeat the request. Nano Banana successfully copes with these tasks.

The AI already knows that we have two images. We tell it to combine them in a certain sequence — done successfully. What anyone can do in Paint is done. 

We are not satisfied with this result. Let’s check if the AI will separate stalker from background and insert it into Snake. Nano Banana takes its pre-edited image without text (we didn’t ask for it) and inserts the character in the middle of the left side. The delta sign in the middle remains. It looks good. Or does it?

We realize that the stalker looks a little smaller than Snake’s head. Let’s try to enlarge the stalker or shrink the Snake’s head. We’ve found something that the AI couldn’t handle. 

We tried to do the opposite or make them the same size — AI does something, but there is no overall difference. The fact that it sees whom exactly it sees for our better understanding proves that everything is fine with contextual understanding. Is the task not done?

Let’s try again. Let’s replace the word “enlarge” with “expand”. It’s more appropriate, since when we edit the image on our own, we’ll be expanding the stalker and cutting out certain parts of it that will be out of frame. This time, Nano Banana is much better. 

Changing the face

We’ll use the Metal Gear Solid Delta: Snake Eater poster with the text cut out. For the experiment, let’s take one of the action heroes — Arnold Schwarzenegger. The “Iron Arnie” looks more like Anthony Starr, but it will do. 

Now let’s try replacing Arnie with the classic Rambo. In addition, let’s replace the black bandage with a red one for more authenticity. The result is good, but it doesn’t look like Sylvester Stallone.

Please add and combine two action heroes into one image. For some reason, the AI does not take Arnold from the previous image (as specified), but draws a new one. Moreover, the new one is much better. In general, you can be quite satisfied with the result.

Scenario for simulating the creation of a poster

When working on a review or article, you often need to make certain posters or photos. Let’s look at a few options that I personally use and see how Nano Banana handles them. For such tasks, I usually do the following: turn on the console, switch to Desktop mode, search for the required image in the browser and open it, take a photo, and edit it in one of the graphic editors. We can do something like this without touching Deck and without physically photos. 

Take one of the previous photos and search for an image for the poster. You can search on your smartphone or computer. You can also make requests for them. One of our future articles will feature the MGS Delta.

It looks good, but you need to change the background for the surroundings. Since part of the game takes place in the jungle, we’ll replace the background with grass. The AI added a lawn. No problem, please change it to much taller grass. This time it’s fine.

Looks good, but it feels a bit empty. Let’s add a random stone and a dog’s footprint. Nano Banana correctly understood the given location of the objects. It has no complaints about the stone, but the shape of the footprint is too artificial. Let’s add more details and get a good poster.

There is a small detail missing — the ITC.ua logo. We took a ready-made one and asked them to put it in the right angle. On the second try, the result is optimal.

Problems during use

The most obvious problem with using AI is not understanding what the user wants. In addition, he or she may formulate requests incorrectly. In the example of “increase in on a stalker” Nano Banana could not understand what exactly was wanted from it. The solution was to change the phrase: increase -> expand. In future situations, try to make it easier for AI to understand what you want.

Censorship should not be forgotten either

The most common technical problem is no image. For example, you upload a photo, write down the necessary steps for processing for a long time. After that, the AI replies “Image is ready”, but it is not in the message. You ask it to show it, and there are two possible outcomes: it does show it or not. This is a clear bug that can be solved with the phrase “Add an image tag”. Since it has already been created, you don’t have to wait long.

AI seems to be playing a joke on us.

The new Nano Banana tool (Google Gemini 2.5 Flash Image) allows you to edit and supplement original images in a very convenient text-based way. For true professionals, it may look like a toy, but for ordinary people it is a real treasure. Yes, there are problems. But treasures are not always in plain sight.