# What is Gemini 2.5 Flash Image, the model nicknamed Nano Banana?

Date: 2025-08-27 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-is-gemini-2-5-flash-image-nano-banana/

**Answer:** Gemini 2.5 Flash Image is Google's new image generation and editing model, released on August 26, 2025, which had been tested anonymously under the nickname "nano-banana." It edits photos from plain-language requests while keeping people, pets and objects looking like themselves, can blend several images into one, and is available in the Gemini app and to developers for about four cents per image.

Yesterday, August 26, 2025, Google released **Gemini 2.5 Flash Image**, a model for creating and editing pictures. For a few weeks, people on LMArena, a public website where users compare AI models without knowing which is which, had been raving about a mysterious image editor listed only as "nano-banana." Google has now confirmed it is the one behind it, and the nickname has stuck.

The model's main promise is consistency. Ask most image tools to change someone's shirt and the face, the background or the pet beside them may quietly change too. Gemini 2.5 Flash Image is designed to change only what you ask for and to keep a person recognizably the same from one edit to the next.

## What Google released

There are two ways in, one for everyday users and one for developers.

| Who | Where | What it costs |
|---|---|---|
| Everyone | The Gemini app, starting yesterday | Included in the app |
| Developers | Gemini API and Google AI Studio (as a preview) | $30 per million output tokens; each image is 1,290 tokens, about $0.039 |
| Businesses | Vertex AI, Google's cloud AI platform | Same model |
| Other platforms | OpenRouter and fal.ai, two developer services | Through their own pricing |

The model comes from Google DeepMind, Google's AI research lab. Google calls it the top-rated image editing model and says it leads LMArena and other benchmarks. For developers, the model is labeled a preview, with a stable version promised in the coming weeks.

## What you can do with it

Google highlights four abilities:

- **Keep a likeness.** Put yourself or your dog in a new outfit, a new place or a different decade, and the face stays the same.
- **Blend photos.** Combine several pictures into one scene, for example a photo of you and a photo of your dog on a basketball court.
- **Edit in several turns.** Keep refining the same image step by step, changing one part while the rest stays put.
- **Transfer a style.** Take the color and texture of one thing, like flower petals, and apply it to another, like a pair of rain boots.

Developers also get targeted fixes in plain words: blur a background, remove a stain from a T-shirt, remove a person from a photo, change a pose or color a black-and-white picture.

## How it works

Many image generators are separate systems: you gave them a text prompt and they painted something new. Gemini 2.5 Flash Image is the image ability built into Gemini 2.5 Flash, one of Google's general AI models. Google calls this **native** image generation: the same model that reads and reasons about your words also produces the picture.

That design matters in two ways. First, the model can draw on Gemini's **world knowledge**, what it has learned about how things work, so it can, for instance, read a hand-drawn diagram and respond to it. Second, it can hold a conversation about an image, which is what makes multi-turn editing possible. Each new request builds on the last result instead of starting from scratch.

Here is a worked example of multi-turn editing, starting from a photo of an empty living room, with what each step should do:

1. "Paint the walls a soft sage green." The walls change; the floor and window stay the same.
2. "Add a tall oak bookshelf on the left wall." The bookshelf appears in the green room.
3. "Put a round coffee table and a gray sofa in the middle." The furniture joins the scene.
4. "Make it evening, with a warm lamp on the bookshelf." The lighting changes while the rest stays.

For a developer, the price works out simply. At $0.039 an image, 100 edited product photos cost about $3.90.

> **Try it:** Choose a photo of a room, a plant or an object you own. Make three small edits in a row, one change per request, and check after each step whether anything you did not ask for changed.

## Safeguards and limits

Every picture made or edited in the Gemini app carries a visible watermark and Google's invisible **SynthID** watermark, a hidden signal that software can detect to show an image was made with AI. Images made through the developer platforms carry SynthID. Google's rules ban creating non-consensual intimate images, and Nicole Brichtova, a product lead at Google DeepMind, told TechCrunch the company wants to give creative control, "but it's not like anything goes."

Google is also clear about weak spots. It is still working on rendering long passages of text inside images, on making character consistency even more reliable, and on getting fine factual details right.

## Why it matters this week

Image tools have become a key battleground among AI companies. TechCrunch recalls that OpenAI's image generator in GPT-4o, launched in March, drove a surge in ChatGPT use. It reports that ChatGPT now has more than 700 million weekly users, while Gemini had 450 million monthly users in July. A popular editing tool is one way Google hopes to close that gap. Meanwhile, Meta announced last week that it will license image models from Midjourney.

For users, the shift is from generating pictures to editing your own. Being able to make careful, step-by-step changes while keeping faces and details intact is useful for everyday tasks, like previewing a room makeover or a product photo, and it also makes realistic fakes easier, which is why watermarks matter.

## How can you edit your own photos with Gemini 2.5 Flash Image?

Open the Gemini app or website, upload a photo, and describe the change you want in a plain sentence. Make one change at a time, and say what should stay the same: "Change only the sofa to dark blue; keep everything else as it is." You can upload two or more photos and ask Gemini to combine them, or ask it to apply the look of one picture to an object in another.

If you are a developer, you can try the model for free in Google AI Studio, where Google offers ready-made example apps you can remix, and call it through the Gemini API under the name `gemini-2.5-flash-image-preview`.

Use photos you have the right to edit, and ask permission before editing pictures of other people. Since the model can make convincing changes, be clear with others when an image has been altered.

## Sources

- Introducing Gemini 2.5 Flash Image, our state-of-the-art image model: https://developers.googleblog.com/en/introducing-gemini-2-5-flash-image/
- Image editing in Gemini just got a major upgrade: https://blog.google/products/gemini/updated-image-editing-model/
- Google Gemini's AI image model gets a 'bananas' upgrade: https://techcrunch.com/2025/08/26/google-geminis-ai-image-model-gets-a-bananas-upgrade/