Nano Banana (Gemini 2.5 Flash Image)

What it is

Nano Banana is the nickname of Gemini 2.5 Flash Image, an image generation and editing model that Google released on 26 August 2025^[1][2]. Where a chat model answers with words, this one takes images together with a written instruction and returns an image. Google described it as the latest upgrade to image generation in the Gemini app^[2].

What it can do

Google lists four capabilities for the model^[1]. The first is blending: several images can be combined into one picture^[1]. The second is character consistency: the model keeps the characters in an image looking like themselves from one edit to the next^[1]. Google said it keeps people and pets consistently looking like themselves across edits^[2]. This is harder than it sounds. Many image models redraw the whole picture each time, so a face or a piece of clothing slowly changes and a long series of edits becomes impossible; holding the subject fixed is what makes repeated editing usable.

The third capability is targeted editing: the change described in an ordinary-language prompt applies to the part of the picture that the prompt is about, rather than to a completely new scene^[1]. The fourth is world knowledge: the model can draw on Gemini's knowledge of the world when it interprets a request^[1], so a prompt can refer to things by name instead of describing every visual detail.

Availability and price

At launch the model was available in the Gemini app and, in preview, to developers through the Gemini API, Google AI Studio and Vertex AI^[1][2]. On the API it was priced at $30 per 1 million output tokens, and since each image used 1,290 output tokens, one image worked out to about $0.039^[1]. Google said that images created or edited in the Gemini app carry a visible watermark and an invisible SynthID digital watermark^[2].

How Google positioned it

Google called it the top-rated image editing model in the world, citing the LMArena image-edit leaderboard^[2]. A leaderboard reflects the votes collected up to a given moment, so the claim is best read as Google's description of the model at the time of release.

Timeline and later use

The August release was a preview for developers^[1][2]. A generally available version, released as gemini-2.5-flash-image, followed on 2 October 2025, with a listed shutdown date of 15 March 2027^[3]. The model also travelled beyond the Gemini app: Google Antigravity shipped with Nano Banana when it launched in November 2025^[4].

Why it matters

The model put instruction-based image editing in two places at once: inside the Gemini app for ordinary users, and inside an API for developers building their own tools^[1][2]. For a user, changing a photograph becomes a matter of describing the change instead of retouching it by hand; for a developer, the same capability can be called from code at a per-image price of a few cents^[1]. The consistency features matter for the same reason: an edit is only useful if the person in the picture is still the same person afterwards, and Google singled out people and pets as the subjects the model is meant to hold steady^[2]. Finally, the watermarks — one visible, one an invisible SynthID mark — mean that pictures made or changed in the Gemini app carry a trace of how they were produced^[2]. The name "nano-banana" stuck alongside the formal model name^[1][2].