Following the buzz surrounding Google's latest generative AI tool, I decided it was time to experience Gemini 2.5 Flash Image for myself. This innovative model enhances your ability to edit photos directly within the Gemini platform. Dubbed the "nano bananas" model by AI enthusiasts, it is inspired by a series of quirky banana-themed teasers released by Google executives. Within just a few weeks of its launch, over 200 million AI images have been created, and more than 10 million users have signed up for the Gemini app, as revealed by Josh Woodward, Google's Vice President of Google Labs and Gemini.
Google has made substantial investments in its generative media models this year, unveiling updated versions of its image and video generators during its annual I/O developers conference. One of the standout features was the AI video generator Veo 3, which showcased synchronized audio—an industry first. Additionally, creators have produced over 100 million AI videos using Google's AI filmmaker tool, Flow.
As someone who has extensively tested various AI creative software, I was eager to see what Google had developed. However, my experience with the Gemini nano bananas model revealed that a flashy introduction doesn't always guarantee performance. Here's a detailed account of my experience, highlighting the good, the bad, and the frustrating aspects of Gemini 2.5 Flash Image.
The Gemini bananas model excels at seamlessly adding elements to existing images, effectively integrating AI-generated components into your photos. One of its standout features is its ability to maintain a consistent character appearance, ensuring that subjects in your photos remain recognizable after editing. Google emphasized its focus on improving these aspects, and my experience confirmed this claim. For instance, in an edited image of my sister and me, our appearances were largely unchanged, demonstrating excellent character consistency.
When I requested Gemini to add a third sister who resembled us, it executed this task remarkably well by placing a similar-looking woman between us. I was also impressed with the speed at which Gemini processed requests, often completing tasks in under 15 seconds. Additionally, all images generated or edited by Gemini included a watermark, providing an essential marker for AI-generated content. Google's SynthID technology further aids in distinguishing AI content from human-created imagery.
Despite its strengths, Gemini bananas has notable limitations. For starters, it automatically generates square images, and my attempts to adapt these to other dimensions often went unanswered or failed to yield results. Furthermore, I noticed a significant reduction in the resolution of many images. Although I typically capture high-quality photos with my iPhone 16, the Gemini model frequently blurred fine details, which is frustrating for any photographer.
In my efforts to use Gemini for complex photo edits—tasks that AI is expected to excel at—I found it struggled with adhering to prompts. For instance, I repeatedly asked it to remove reflections from a photo of a Freaky Friday movie poster, but the reflections persisted. Each attempt only degraded the image quality further, resulting in illegible text and distorted faces of the movie's stars, Lindsay Lohan and Jamie Lee Curtis.
Additionally, Gemini nano bananas faced challenges with resizing and cropping images, which are fundamental aspects of photo editing. I reached out to Google regarding these resolution and dimension issues, and a spokesperson acknowledged the challenges, stating that the company is actively working on improvements.
Overall, my experience with Gemini nano bananas demonstrated Google's commitment to remaining a frontrunner in generative media. However, significant drawbacks exist, particularly its emphasis on creating new elements rather than refining common photo editing tasks. Currently, the nano bananas model is best suited for users who prioritize speed and major edits over precision. For those seeking more detailed editing tools, it may be wise to explore other options or await future updates from Google.
Accessing the new model is straightforward as it is automatically integrated into the base Gemini 2.5 Flash model. Gemini is available for free, with options for more models and higher usage caps through Google's AI plans, starting at $20 per month. Paying subscribers may also access the model through Google AI Studio. Users simply need to upload an image and provide a prompt, with each request consuming between one to two thousand tokens, depending on the level of detail required.
Moreover, users of Adobe Express and Firefly can now access the new model as well. It's important to note that Google's privacy policy indicates that uploaded content may be used to enhance its AI products. Therefore, it is advisable to avoid including sensitive or private information in your uploads. The company's prohibitive use policy also strictly prevents the creation of illegal or abusive material.
