In this video I tested a new AI image model called “Nano Banana” (a codename a lot of people think is Google), and I go through why some folks believe it’s basically Gemini 3.0 in disguise. I show community examples first—stuff like changing lighting and tone, adding/removing objects, turning two separate photos into a believable selfie, and doing clean product-style edits. The big theme: when Nano Banana is in its element, it’s insanely good at faithful edits that keep the original style intact. Then I walk you through how to try it right now on LM Arena (new chat → battle → generate images), plus the catch: you won’t get Nano Banana every time. In my testing it showed up roughly 20–25% of the time, so you have to run a few battles. My best results came from image-to-image editing: it nailed small, specific changes (like only changing water color, or editing feathers without touching the whole bird) and it even handled adding text pretty well while matching the image’s color scheme. The main weakness I saw is photorealism and “convert this illustration into a realistic photo” requests—often it barely changes the image or just doesn’t follow the realism instruction. My takeaway: upload a reference image and edit from there for the most consistent wins.