I tested Google’s latest Nano Banana Pro, and it surprisingly removed my clothing. I didn’t request it to, but the AI model clearly believed my greeting card would appear better with more bare skin.
As indicated by its name, Nano Banana Pro targets professionals. Powered by Gemini 3, it represents an enhancement of the company’s well-received image generation and editing tool that gained popularity during a social media phenomenon which converted selfies into hyperrealistic 3D models. Google claims it enables the creation of high-quality images suitable for printing, applying readable text to images, and merging various pictures into a single work. It’s also intended for “individuals looking to feel like professionals,” according to Naina Raisinghani, a product manager at Google DeepMind, who spoke to The Verge. This sounds promising, as I certainly don’t consider myself a professional. The outcomes for me were shiny, yet silly. They appeared good but felt unpolished.
Using Nano Banana Pro is quite straightforward: enter the Gemini application, choose “create images,” and activate the ‘thinking’ mode. Just enter your prompt (and picture, if applicable) and proceed. It’s also free, although there are restrictions, with quotas increasing for Google AI Plus, Pro, and Ultra users.
Google makes bold assertions, boasting “studio-quality designs,” “impeccable text rendering,” and a variety of creative editing features. To evaluate these promises, I uploaded a simple snapshot of myself outside The Verge’s office in New York with the Brooklyn Bridge in view. I instructed Gemini to alter the lighting from daytime to nighttime, and it performed admirably. The output seemed credible. It even managed to address details that frequently challenge image generators, such as ensuring cars moved in the correct direction. Adjusting the camera perspective was just as simple. I requested Gemini to replicate the image as if it had been captured from an elevated angle on the right, and it succeeded.
Google further claims that Nano Banana Pro can produce infographics and diagrams to represent real-time data like weather or sports. Being from Britain, I inquired about the forecast for the next four days in Washington, DC, and New York City, where I am currently located. Visually, the infographic would have fit seamlessly on a basic weather site. The text and figures looked normal — a significant improvement over the jumbled words you often find in AI-generated images — and Gemini provided a list of sources at the conclusion that assisted me in verifying its accuracy.
The model encountered some difficulties with more elaborate tasks. I asked it to condense a recent Verge article about how Europe is moderating its AI and privacy regulations into a comic book-style interpretation. The visuals and text were rendered nicely in a cartoonish typeface, but the comic did not summarize the story, instead offering a vague description of the EU’s AI Act. The issue may have stemmed from the fact that I provided Gemini with a link to the story instead of pasting the content.
However, it produced a satisfactory comic-style summary when I did paste the text. It conveyed the essence of the actual article, although I doubt I would have fully comprehended it if I hadn’t authored the source material. There were also fabricated phrases included that had not been present in my article.
To truly feel like a professional designer, I attempted to create greeting cards. Christmas is approaching, after all. Given that I merely uploaded three selfies, Gemini astonishingly produced three full-body renditions of me, each displaying different outfits and expressions. It also crafted a convincing, snowy backdrop with Christmas trees as I requested and displayed “Merry Christmas!” at the top, just as I asked.
Gemini decided to take creative liberties when I requested it to swap the snowy scene for a summer beach setting for an Australian-style holiday. Those liberties included my deepfaked outfits: two of my clones ended up shirtless. It was peculiar. There were also some conspicuous AI-generated feet and a cheerful sandman to replace the snowman from the winter scene (being constructed by my shirtless double). Some problems did arise, however — the sandman lacked a shadow, unlike the other rendered elements in the image, and the Christmas lights in the palm trees had a strange glow under the bright sunlight. I tested its precision editing capabilities by asking it to enhance only one clone’s physique, which it accomplished in moments (if only it were that simple in reality). Overall, the quality was outstanding, and the image could have been somewhat plausible (aside from the abs) had there not been a significant tattoo missing from my chest.
It wasn’t perfect, though. The model struggled with maintaining the original text on my card that I requested. Instead of “Merry Christmas!” it chose to display “Aussie Summer Christmas!” It also seems to face challenges with animals: my sister’s cat sat in the same awkward pose as in the reference image I provided in every version of the card (though he was donned with a playful Santa hat).
Overall, I found it impressive. Nano Banana Pro is a significant improvement over the basic model. I could request more specific alterations, and it actually generated coherent text, eliminating a major hurdle preventing generative AI tools like this from being applicable in the real world. However, these features still fell short of transforming me into an exceptional designer.
