Ready for another bleak application of generative AI? I’m not sure I am, but here we are with Gemini’s ‘Nano Banana’. Alright, alright, get it out of your system and let me explain. The aforementioned fun-size fruit is in fact the name of a major update to Gemini’s AI image generation capabilities. The update is also known as Gemini 2.5 Flash, but I’m going to keep calling it Nano Banana in an attempt to stave off the existential dread about deepfakes.
Back in April, Google introduced native image editing to the Gemini app. Nano Banana upgrades this image editing model, meaning the app can now generate visual variations on a character or image subject with frankly alarming consistency. I would personally advise against feeding your likeness into any genAI model, but if you’ve always wondered what you’d look like as a matador, a 90s sitcom character, or even, heck, an artist, Google demonstrates how the Gemini App can now realise that vision.
Google writes, “Gemini lets you combine photos to put yourself in a picture with your pet, change the background of a room to preview new wallpaper or place yourself anywhere in the world you can imagine—all while keeping you, you.” The blog post also highlights that this update is currently sitting at the top of LMArena’s image editing model leaderboard, so allow me to explain what all the hype is about.
Besides smooshing subjects from multiple photos together, giving them all costume changes, and switching up the scenery, the Gemini app now offers multi-turn editing as well. Snap a picture of an empty room, and then task Gemini with filling it without finding yourself transported to somewhere totally different over the course of a few text prompts. It’s the ability to do these small, incremental edits without noticeable inconsistencies creeping in is what’s chiefly turning heads.
It’s smart, for sure, but what I will say is that I’ve still not managed to get it to effectively crop a square image to 16:9, and that is real basic editing stuff, Googs. Gemini just told me it was “not capable of performing precise edits like cropping an image to a specific aspect ratio.”
Sure, you can still use the Gemini app to create fun selfies or do a spot of interior decorating, but you can also use it to manipulate the image of dead celebrities to create completely ahistorical images. I used to think I was okay at identifying the ‘sheen’ of AI generated imagery, but taking Microsoft’s Real or Not test last month and now this demonstrates that I was vastly overestimating my ability.
To be fair, all images generated with the Gemini app have a visible watermark on them to signify that they’re AI—but this is subtle enough to either be glanced past or for some bad actors to simply crop out in other image editing software. Google writes that all images also additionally sport an “invisible SynthID digital watermark” which should be picked up by the company’s own SynthID Detector. However, this is not yet widely accessible, as the program is still looking for both partners and early testers.
Frankly, Nano Banana and big tech’s ‘move fast and break things’ approach to genAI makes me want to never put my face out there on the internet again.
1. Best overall:
HP Omen 35L
2. Best budget:
Lenovo Legion Tower 5i
3. Best high-end:
Corsair Vengeance A7500
4. Best compact:
Velocity Micro Raptor ES40
5. Alienware:
Alienware Area-51
6. Best mini PC:
Minisforum AtomMan G7 PT
