Nano Banana AI: How and why did Google come up with this weird name?

Google’s viral AI model, Nano Banana, gained fame for its quirky name. But the story behind it reveals an unexpected twist from inside Google.

Google’s new image generation model, known widely as Gemini Nano Banana, has drawn attention for both its name and its performance. However, few knew how the name came about until David Sharon, Group Product Manager for the Gemini App, explained the story on the Made by Google podcast.

Google’s viral AI model “Nano Banana” got its quirky name by accident.(AI-generated)

The Origin of the “Nano Banana” Name

According to Sharon, the name “Nano Banana” began as a placeholder. An employee named Nina submitted the model anonymously to LM Arena, a public platform where users test AI models without knowing their source. To maintain anonymity, she entered a temporary name at random. “Nano Banana was created by a PM named Nina,” Sharon said. “When you submit a model anonymously to LM Arena, you must include a placeholder name.”

He added that Nina came up with “Nano Banana” at around 2:30 a.m., and the quirky name unexpectedly caught on. Google’s team intended to test the model like any other, without revealing that it came from the company. But as soon as users began interacting with it, the placeholder name gained popularity on the platform. “People were already calling it Nano Banana, so we decided to keep the name,” Sharon said.

Why Nano Banana Stands Out

Beyond the name, Gemini Nano Banana offers advancements in image generation, particularly in maintaining subject identity and creating complex, imaginative visuals. Sharon explained that the model differs from its predecessors by ensuring consistency in generated characters. “When I first tested Nano Banana, I uploaded an image of myself and asked to place myself in space. For the first time, I saw myself, not a distorted version,” he said.

This consistency, Sharon noted, became a central focus of development. Users want to see familiar faces, such as themselves, family members, or pets, accurately portrayed in creative settings. The model’s strength lies in its capacity to preserve visual identity while meeting imaginative requests.

An internal “Greenfield” team at Google tested Nano Banana through multiple creative challenges. They explored its conceptual blending ability, such as combining unrelated objects into one meaningful image. For example, when asked to merge a couch and a potato, the model successfully generated a “couch potato” image.

These results showed that Nano Banana extends beyond standard image generation, which handles abstract and conceptual ideas effectively. Its success in creating realistic yet inventive visuals demonstrates the evolving potential of AI image models in creative applications.