What Was Nano Banana? Google Image Model Guide

What Was Nano Banana? Google's AI Image Name Explained

Nano Banana began as the public codename for Google's Gemini 2.5 Flash Image model before its August 2025 launch. The name later became a public brand for a growing family of Gemini image-generation and editing models.

That means "Nano Banana" can describe a specific historical model or the broader family, depending on the date. A reliable comparison needs the exact model identifier.

The short timeline

flowchart LR
    A["LMArena public codename"] --> B["Nano Banana"]
    B --> C["Gemini 2.5 Flash Image"]
    C --> D["Nano Banana Pro"]
    C --> E["Nano Banana 2"]
    C --> F["Nano Banana 2 Lite"]
    C --> G["Legacy original"]

The name was not a leaked internal secret

Venture Step E082 described Nano Banana as an internal Google codename that leaked and became more popular than the official model name. Google later published a first-person origin account that makes the story more specific.

The team needed a public codename for an anonymous LMArena submission. Product manager Naina Raisinghani proposed Nano Banana during a late-night exchange. She said friends called her "Naina Banana" and "Nano," so she combined the two. Google introduced the model under that name on LMArena in early August. Google's Nano Banana name history

The name therefore did conceal the model's identity during public evaluation, but it was itself public. The transcript remains preserved as the episode-era source. The derived page uses Google's later account rather than repeating the leak claim.

What Google launched in August 2025

On August 26, Google announced a new image-editing model inside the Gemini app. The company emphasized maintaining a person or pet's likeness, combining multiple photos, transferring styles, and continuing an edit across several conversational turns. Google's launch announcement

The technical name was Gemini 2.5 Flash Image. Google described Nano Banana as the latest upgrade to image generation in Gemini, while the developer model later used the identifier gemini-2.5-flash-image.

Google's launch page also said images created or edited in the Gemini app included a visible watermark and an invisible SynthID watermark. That statement belongs to the launch-era Gemini app. Current interfaces, account tiers, API outputs, and watermark presentation need their own dated checks.

Why version names matter

The original model was optimized for speed and efficiency. The current Google developer page still identifies Gemini 2.5 Flash Image as Nano Banana and lists it for high-volume, conversational editing and low-latency workflows. It also lists the preview version as deprecated. Gemini 2.5 Flash Image documentation

Nano Banana Pro is a different model. Google launched it in November 2025 as Gemini 3 Pro Image, with more emphasis on complex instructions, knowledge-informed visuals, text rendering, and output up to 4K. Nano Banana Pro announcement

By July 2026, Google's image-generation guide described four choices:

Google's current guide recommends a newer model for many new workflows and calls the original the legacy pioneer of the series. Current Gemini image-generation guide

That current family state should not be inserted into a 2025 test. E082 tested the original launch-era experience as Dalton understood it. A 2026 test would be a new evaluation.

Model, feature, app, and brand are different layers

Nano Banana has been used in several ways.

At launch, it named the model and editing upgrade people encountered in Gemini. Later, Google used the name in Search, NotebookLM, and Photos. The company said more than five billion images had been generated by October 2025, then expanded the capability into products where people already worked with images. That is a Google-reported usage figure, not independent evidence of retention or market leadership. Google's October 2025 expansion record

The same brand can therefore sit above different surfaces. A Photos user may encounter a template. A Gemini user may have a conversation. A developer chooses a model identifier. A professional workflow may route to Pro.

When documenting a result, record the product surface and the model separately.

What the name does not establish

The name does not prove that the model caused an app-store ranking, displaced ChatGPT in durable use, or led every image-editing task. It also does not tell a current user which model handled a request unless the interface or API exposes that information.

The episode title called Nano Banana the trend that "dethroned ChatGPT." No dated ranking record was recovered in the E082 package, and a temporary store position would not establish general market leadership.

The durable product story is simpler. A strange public codename attached itself to a capability people could demonstrate visually. Google then adopted the name and extended it across later products and models.

For the original experiment, see [[Venture Step E082 - Testing Google Nano Banana Live]]. For a current reusable entity record, see [[Nano Banana Product Profile]].

AI assistance was used for research organization, drafting, and validation. Publication remains unauthorized.