Meet Nano Banana: Google’s Image Model That Actually Works
A few weeks ago, I took a beautiful headshot — but embarrassingly, I showed up to the shoot in a casual, fun graphic T-shirt. Not exactly the most professional choice. I’m sure I’m not the only one who’s been there, right? 😅
Luckily, I have data centers working overtime on my behalf, which means I was able to transform my great-but-imperfect headshot into endless polished options with just a few clicks.
If you’ve used image models before, you know they often struggle. They glitch on multi-edit sequences and produce a glossy, fake look that just feels off. But Google’s new image model changes that.
To highlight the difference, I tested the same prompts with both ChatGPT’s image model and Google’s Gemini 2.5 (a.k.a. Nano Banana). Spoiler: the results weren’t even close.
Here is the initial prompt:
As you can see, Gemini’s output is clearly better, even though I look dashing in both. Now, the next prompt is where the difference is clearly evident.
Prompt: Add a black filter to this image
Not much needs to be said, but ChatGPT transfigured me into a nightmare.
Next Prompt: Update my clothing to be a tan quarter zip
There is a stark difference in the output.
As we can see, Gemini's new image model is amazing with multi-turn editing.
Beyond making me look more professional (and, honestly, a little more handsome), the use cases are expanding quickly. The internet is buzzing with creative applications, and some of them are game-changing.
You should definitely check this X thread from Google Gemini on the crazy use cases: https://x.com/GeminiApp/status/1961192420496085127
Beyond the amazing output from Gemini, Gemini's integration of SynthID is very important as we encroach into a world where deepfakes become more common. SynthID makes every Gemini-generated image with visible and invisible watermarks (via SynthID).
The future is both exciting and scary, but it’s definitely fun playing around with the new technologies that emerge. If you’ve played around with the new image models, I would love to see some of the images you generated.