Google’s Nano Banana Pro creates ultra-realistic images using artificial intelligence. It scares the hell out of me


One of the most important things I do when testing new AI-powered photo and video generators is try to figure out whether the claims a company makes are true or just marketing nonsense. Google promised Nano Banana Pro Gemini 3 would bring understanding of the world and the informational power of Google search to your AI photos. I jumped on the new model as soon as it was released, and frankly, I think Google is underestimating its capabilities.

the The original nano banana The model was very popular, although it exists like any AI service Stark areas For improvement. There’s one area I’m really impressed with the new Pro version: it can really handle displaying complex text in images. Nanobanana pro can create complex graphs and add captions to images without much of the usual hallucinations. This changes the rules of the game, but perhaps not in the way we hope. Throughout my initial testing with the pros, this was a constant feeling. The Pro model is excellent, albeit in a very worrying way.

Here’s what my initial testing showed, complete with several photos of the nanobanana pro. If you want to try it for yourself, here it is Cheapest way to get therealong with The new image detector with artificial intelligence technology from Google.

The people and characters are super realistic

The TV show Riverdale has given us a lot of silly and funny dialogue, but nothing is funnier to me than Archie’s line about the “epic highs and lows of high school football.” As someone whose high school football team was a three-time state champion, I could see where Archie was coming from. Nanobanana pro was able to recreate that moment.

AI photo of KJ Apa wearing a varsity jacket, with caption: It means you never know the wins and defeats, the epic highs and lows of high school football

This is a perfect-looking photo of KJ Apa, who played Archie, although he’s not as into the action as he was in the original scene.

Created by Katelyn Chedraoui using Gemini AI

Gemini initially rejected my request when I used the names “KJ Apa” and “Riverdale” in my claim (just as they should), but spit out this very accurate-looking photo after I simply deleted the names. This didn’t always work — Gemini didn’t flag my request for a scene similar to Legally Blonde, but he didn’t pick up on Reese Witherspoon’s distinct resemblance, either. But I did get two accurate depictions of Dr. Seuss’ The Grinch: Jim Carrey’s version and the cartoon version voiced by Benedict Cumberbatch.

These two side-by-side images of The Grinch were generated by artificial intelligence.

Dr. Seuss is probably turning in his grave right now.

Created by Katelyn Chedraoui using Gemini AI

Although these results are impressive, it is not a good thing for a Gemini to be able to achieve them. Celebrities, public figures, and other people known to them He expressed his concerns About how AI tools are making it easier for anyone to create realistic-looking AI versions of themselves. AI companies said there are technical barriers and policies in place to prevent abuse, but we’ve already seen how these systems can fail. As new models like the nanobanana pro become more capable of creating these images and videos, it’s becoming harder to tell if something is real or artificial intelligence.

Display ridiculously accurate text in images

Store signs, T-shirt logos, and anything that requires legible text have long plagued AI image generators. These errors have been useful to us when looking for signs of whether something was created using AI, because the text will always be sloppy. But the Nano Banana Pro solves that problem almost entirely, for better or worse.

AI companies have long aimed to create readable text in their generative media. OpenAI’s first image model — which you may remember spurred a trend of people making copies of Studio Ghibli for themselves — was supposed to excel at generating text in images. But my testing showed that there are still major problems with this.

Nanobanana pro does not have any obvious errors in its graphs. Moreover, their designs are clean, well-organized, and efficient. Check out some of the ones I created.

An AI-generated infographic explaining the water cycle.

ChatGPT actually suffered from typos and fabricated fake stages of the water cycle. Nano Banana Pro excelled.

Created by Katelyn Chedraoui using Gemini AI

The AI-generated graph details three separate impacts. Systemic damage, what's in a cigarette, and statistics.

Nanobanana pro relies on sources from Google search to create graphs.

Created by Katelyn Chedraoui using Gemini AI

AI-Generated Image explains how AI generators work in three elegant steps. The chart includes a quote.

Nanobanana pro infographics can analyze complex topics, as in this infographic on how AI-powered image generators work.

Created by Katelyn Chedraoui using Gemini AI

Nanobanana pro was created using Gemini 3 and relies on Google’s massive search database of online content to create these graphs. But if you don’t want to let Google choose the source, you can include the sources you want to pull it from in the claim. I gave Gemini my link In-depth reporting on Hollywood’s power struggle with artificial intelligence To test this out, and I have to admit, it’s a really good visual representation of the story.

An AI-generated image that pits human creativity against algorithmic efficiency.

This is exactly what an infographic should do by condensing the story into its main themes and connecting them. Additionally, it included the (old) CNET logo and source as you requested.

Created by Katelyn Chedraoui using Gemini AI

Nanobanana pro is alarmingly excellent

There have been many watershed moments in the development of generative AI. We’ve seen several of them in just the past six months, with Veo 3 introducing audio in AI-powered videos and Sora making deepfakes easier than ever before. The original Banana Nano model was another one, but the Pro model clearly outperforms anything we’ve seen in AI photo technology this year or any year before. These nanobanana pro AI images are the closest I’ve ever seen to human-created designs and photography. While I’m very impressed with the technical prowess that Google displays, it only heightens my concerns about the misuse of AI technology like this.

I tried to convince Gemini to make infographics about questionable pseudoscience theories. It stopped any requests that were (correctly) classified as false health information, although it did allow one photo to pass through — showing that the guardrails in place are still not complete. We’ve already seen how these tools can be used to create racist, offensive, and illegal content, despite the efforts of AI companies. Nanobanana pro can load the bad actors’ arsenal with all the firepower they need. I’m unfortunately sure it will only be a matter of time before we see AI images generated by nano-bananas attached to viral social media posts or campaigns, with no way for us to determine with our naked eyes whether they are real or AI.

At the same time, since nanobanana pro is the most AI-capable image model available, it will be very useful. People who want a quick image, help editing existing images, or have an infographic to include in their work will be able to create high-quality results. This is the double-edged sword inherent in generative media innovation, and Nano Banana Pro is the sharpest tool – or weapon – yet.



Leave a Reply

Your email address will not be published. Required fields are marked *