It started like most internet crazes do these days: a small, clever tweak to an app and a stack of copy-and-paste prompts. Two weeks after Google quietly dropped an image-editing model inside Gemini, people began turning selfies and dog photos into uncanny, lifelike mini-figurines — tiny desktop dolls complete with packaging and mock wireframe previews — and the rest of the internet sprinted to try it. What looked like a novelty quickly became a growth engine: downloads spiked, the Gemini app climbed app-store charts, and Google engineers found themselves triaging what they called a “full-on stampede.”

Meet Nano Banana (aka Gemini’s image secret sauce)

Internally nicknamed Nano Banana, the model launched as part of Gemini’s image capabilities in late August and is marketed as a “transform images in amazing new ways” tool inside the Gemini app. In practice, it’s a lightning-fast image editor: you drop in a photo, tell Gemini what you want changed — from redecorating a room to trying a 1960s beehive — and the model outputs a consistent, editable new image. It can combine multiple pictures, preserve likeness across edits, and even generate short, playful videos by feeding edited frames back into Gemini. The result is a highly polished, iterative image workflow built right into a chat/assistant experience.

That polish is important. Where many image models produce one convincing shot and then break character on follow-ups, Nano Banana appears to keep subject likeness and styling steady across multiple rounds of edits — which makes it much easier for users to create a believable “product” shot of a 3D figurine based on one or more photos.

Numbers that look like a mini-viral cyclone

Google hasn’t issued a traditional press release with headline numbers; instead the company’s Gemini team shared metrics publicly on X. Over the span of a few days, Gemini’s vice president Josh Woodward posted that Nano Banana had driven tens of millions of first-time users to the app and that hundreds of millions of image transforms had been processed — an update that read like a growth startup’s weekly deck: millions of new users, hundreds of millions of images.

Those raw totals translated into chart movement. As of mid-September, the Gemini app was the top free app on Apple’s App Store in several major markets — including the U.S., U.K., Canada, France, Australia, Germany and Italy — frequently nudging OpenAI’s ChatGPT into second place. For a company that has been building a slow, broad AI play for years, that kind of chart momentum is a rare, concentrated win.

The figurine moment — why this prompt caught fire

There are a lot of fun edits people can do with Nano Banana — vintage portraits, anime avatars, fantasy skins — but the runaway hit has been the 3D figurine: give Gemini a selfie (or a few pictures), include a prompt asking for a small desktop figurine with a display box and a wireframe CAD screen behind it, and the model renders a convincing product-photo-style image. The trick is prompt craft: a handful of highly replicable phrasing choices produced consistent results, which made the meme easy to copy across Instagram, Threads, Discord and X. Google even shared example prompts to help keep the trend circulating.

That replicability is what makes a trend stick. The figurines look cute, they’re shareable, and they tap two modern obsessions at once: personalized digital vanity (what would a tiny, boxed me look like?) and the aesthetics of consumer product photography. Once creators and influencers start posting side-by-side before/after images, algorithmic feeds do the rest.

Engineers and limits: “TPUs red hot”

Viral demand created operational headaches. Woodward’s posts candidly described Google’s internal strain — think pagers, hot TPUs, and temporary usage limits while the team scaled capacity — language that undercuts a common myth: that tech giants never scramble. “It’s a full-on stampede to use the Gemini app,” he wrote, noting that in some places Google had to institute temporary caps to keep the service stable. Those engineering details reinforced that this was not just a playful niche; it was sudden, high-throughput demand on real infrastructure.

Culture, commerce and caution

The trend has also produced a local color effect: politicians in Goa, hobbyists in Mumbai, and across India, large numbers of new users jumped in, prompting advisory stories from local outlets and broader conversations about safety and privacy. Experts and some newsrooms have begun flagging concerns: uploading unblurred, identifying photos to AI services raises questions about how images are stored, whether models retain facial data, and how generated outputs could be reused. Some police and civic pages have issued cautionary notes about uploading sensitive images. The fun is real, but so are the stakes.

There’s also a commercial angle. Highly shareable image features drive installs and engagement — which in turn make an app more attractive as a front end for other services, ads or pro tiers. For Google, a user who now opens Gemini to make an avatar is one click closer to other Google products and experiences. For competitors, the lesson is simple: generative AI wins that feel native and social are sticky.

What this means for the AI landscape

Is this the moment Gemini leaps past ChatGPT permanently? Not necessarily. ChatGPT still dominates in conversational breadth and developer integrations, but Nano Banana shows a path to rapid mainstream adoption: pick a delightful, visual use case, make it dead simple, and let social platforms amplify the result. In that sense, the figurine frenzy is not just a meme — it’s an example of product-led virality that can change market share overnight if the company can keep the service running and handle the inevitable scrutiny.

The long tail: creativity, copycats and regulation

Expect copycats and creative forks. Image models are already a hot category, and once a format proves viral, smaller startups and big rivals alike rush to replicate it. At the same time, policymakers and privacy advocates are watching. Viral features that depend on personal likenesses will accelerate debates around consent, data retention, and the boundaries between playful image edits and potentially harmful image uses.

For now, Nano Banana’s figurines are a cultural toy: clever, uncanny and eminently shareable. For Google, they’re also evidence that the right little model — well-integrated into an existing product and handed to a social audience with an easy prompt — can deliver the sort of breakout moment companies chase for years. Whether it becomes a durable competitive advantage depends on whether Gemini can keep the lights on, stay on the right side of safety conversations, and keep giving users reasons to come back beyond a single viral trick.