Exploring Nano Banana: Our Hands-On Experience with Google's Flash 2.5 Image Generation

Nano Banana, Google's Gemini 2.5 Flash image generation tool, has been making waves with its impressive capabilities. We decided to put it through its paces with some real-world editing scenarios to see how it performs, using basic usage. Here's what we discovered when testing four specific use cases.

What is Nano Banana?

Nano Banana is essentially the image generation component of Google's Gemini 2.5 Flash AI model. It allows users to create and modify images through simple text prompts. The technology can understand complex instructions and attempt to make precise edits to existing photos - from removing people to swapping positions and even changing clothes.

How we used it

We used simple usage, where didn’t use any prompt engineering, just barked commands to Google Gemini through the portal and awaiting his feedback.

Simple usage of GenAI as we see it at Agoya

For our test, we uploaded a team selfie we took when we started making some online content featuring Marco (left), Yasmine (middle), and myself (right)

The selfie of the team used for the Use-cases

Use case 1: Removing a person and repositioning

We asked Nano Banana to remove Yasmine and move Marco and me closer together using the following prompt:

Can you remove Yasmine from the picture, and move Marco and myself closer together?

Gemini generated picture after removing Yasmine from the picture

The AI executed the first part flawlessly, Yasmine was cleanly removed from the image. However, when we asked it to close the gap between Marco and myself completely, it struggled. using the following question (Yes i still use Thank you, and please):

Thanks for removing Yasmine, can you now make sure there is no gap between Marco and Myself and it looks like we are standing together?

Despite multiple attempts with different phrasings, the AI couldn't quite get us positioned naturally together without the gap.

Use case 2: Replacing a person

Next, we challenged Nano Banana to replace Yasmine with our colleague Cyriel. After providing some reference photos of Cyriel, the AI generated an image where someone new was indeed inserted in Yasmine's place.

Gemini generated picture of "Cyriel"

The interesting observation here was that while the AI perfectly removed Yasmine and added someone else, the inserted person didn't actually resemble Cyriel at all - despite having reference images to work from. It succeeded at the replacement task technically, but not at creating an accurate likeness.

Use Case 3: Swapping Positions

Our third test involved asking the AI to swap the positions of Marco and myself in the original photo. This proved challenging for Nano Banana.

In the first attempt, it only partially completed the task - placing Marco on the right but failing to move me to the left. When we pointed this out, it tried again and executed a position swap that looked kinda natural and maintained our original appearances.

Use Case 4: Clothing Swap

Finally, we attempted what was perhaps the most complex editing task - swapping clothes between Marco and myself. We wanted Marco to wear my black t-shirt while putting me in his white polo.

This proved to be beyond Nano Banana's current capabilities. Despite multiple attempts with increasingly specific instructions, the AI simply returned the original image repeatedly without making the requested changes.

Conclusion

Our experiments with Nano Banana revealed both impressive capabilities and clear limitations. The technology excels at removing elements from images and can sometimes perform basic replacements, think about a dog or a kid photo-bombing your pictures. However, it struggles with more complex spatial manipulations like closing gaps between subjects, swapping positions naturally, or performing clothing exchanges.

What's particularly interesting is that the AI sometimes believes it has completed a task when it hasn't, returning essentially unmodified images while indicating success. This suggests that while the text understanding is strong, the visual manipulation capabilities still have significant room for improvement.

For simple edits like removing unwanted elements from photos, Nano Banana shows promise and does what it has to do. But for more nuanced editing tasks, traditional photo editing tools still maintain a clear advantage - at least for now. As with all AI tools, we expect rapid improvements, so it's worth keeping an eye on how these capabilities evolve.