Google Nano Banana 2: Promises and Limits of Image AI

Google Unveils Nano Banana 2: A Mixed Advancement in Image AI

Three months after the launch of its latest image AI model, Google is back in the spotlight with Nano Banana 2, unveiled at the end of February. This second-generation model promises to maintain the speed of its predecessor while integrating the advanced knowledge of Gemini 3 to deliver more realistic results. As an expert in creative AI, I put this new tool to the test to evaluate its performance. While Nano Banana Pro remains our preferred choice among image AI tools, this new model raises questions about the potential overperformance of current AI tools.

After testing Nano Banana 2, it appears that Google has partially delivered on its promises. The model proved to be fast, confirming Google's position as a key player in the AI field. However, gaps remain, indicating that improvements are still needed. Here’s how Nano Banana 2 compares to previous versions, particularly in the area of photo editing, which is one of its main use cases.

Photo Editing: Uneven Performance

During my tests, I challenged Nano Banana 2 with a significant task: correcting a photo from the poster of the movie Freakier Friday taken last year, where the evening light created an annoying glare on the glass. Neither Nano Banana 1 nor the pro model managed to solve this issue. I hoped that Nano Banana 2 would rise to the challenge, but it did not.

The AI failed to eliminate the glare, regardless of the model used. This highlights a notable weakness, as one of the major selling points of creative AIs is their ability to correct human errors in photos. In this case, the AI did not live up to that promise.

In less complex scenarios, Nano Banana 2 showed better performance. For example, it successfully removed minor reflections on the glasses of Allyza Umali, Senior Social Media Manager at CNET.

Speed vs. Accuracy

One of the obvious advantages of Nano Banana 2 is its speed. While the pro model takes between 2 and 5 minutes to generate an image, the base model in fast mode produces results in under a minute. However, this speed sometimes comes at the expense of accuracy.

In one test, I asked Nano Banana 2 to place a family photo on the football field of Appalachian State University in North Carolina. While the model did a decent job, the pro model provided finer details and added readable text, such as the name of the stadium, without explicit request.

Style and Photorealism

For stylistic transformations, such as adopting the Pointillism style, Nano Banana 2 and the pro model showed similar performance. Upon zooming in, it can be observed that the pro model produces finer dots, but the difference is not significant.

Nano Banana is renowned for its photorealism, making it a preferred choice for those seeking AI-powered photo editing. However, it is sometimes necessary to accept that certain photographic errors cannot be corrected, and that our photos retain their charm despite their imperfections.

Choosing the Right Model

Nano Banana models dominate the market for good reason. They excel in various styles and outperform other tools in terms of photorealism and text inclusion. While they are not perfect, they are worth exploring for those seeking effective AI options.

Since the launch of Nano Banana 2, Google has removed the original model, leaving the choice between the latest version and the pro model. For quick ideation needs where details are not crucial, Nano Banana 2 is sufficient. In contrast, for work requiring greater precision, the pro model is more suitable, although one must be patient for the results.

The rise of image AI tools, with Nano Banana leading the way, raises concerns among artists about the impact of AI on creativity. While some creators embrace this technology, others express reservations, raising legal, ethical, and environmental questions.