If you have spent any time around AI image tools in the last year, you have probably seen the phrase “Nano Banana” in a prompt thread, a tutorial, or a social media post. It sounds like a joke name, and in a way it was. Nano Banana started life as a secret codename Google used while it tested a new image model in public. People noticed the mystery model’s unusually good edits, the name leaked, and it stuck. Today, Nano Banana is the informal name for a whole family of Google image generation and editing models — and as of late October 2026, the version you should actually use is Nano Banana 2.1.
This guide explains what Nano Banana is, how it works, what it can do, where you can use it, and what it costs. It also clears up the version confusion, because the naming has changed twice and one older version is about to stop working entirely.
The Short Answer
Nano Banana is Google’s AI image editing model family, built by Google DeepMind. You describe the change you want in plain words — “put this person in a rainy Tokyo street,” “change the dress to red,” “turn this photo into a watercolor” — and the model edits the image while keeping the subject looking like itself. That last part is the whole point. Most image editors either regenerate the whole picture (and lose the likeness) or make clumsy local patches. Nano Banana was designed to hold on to a person, pet, or product across edits, scenes, and styles.
The name’s history runs like this. In mid-2025, Google quietly tested the model under the codename “Nano Banana.” It was formally released as Gemini 2.5 Flash Image in the Gemini app and for developers. A later generation, Nano Banana 2 (model id gemini-3.1-flash-image), followed. Then, on 6 October 2026, Google released Nano Banana 2.1 (model id gemini-nano-banana-2.1) as the current version. If you are starting today, Nano Banana 2.1 is the model that matters.
Why Nano Banana Got So Much Attention
Before Nano Banana, AI image editing had a frustrating trade-off. You could generate a beautiful image from scratch, but editing a real photo usually meant one of two disappointments: the edit changed the person’s face, or the edit only worked if you accepted a completely new image that merely resembled the original. Creators who needed the same character in ten different scenes — a children’s book, a product campaign, a comic — had to fight the tool on every frame.
Nano Banana changed the conversation because it treated consistency as the core feature rather than a bonus. You could keep the same character’s face, the same product’s shape, or the same pet’s markings while changing the outfit, the background, the lighting, or the art style. Multi-turn editing made it practical: you could make one change, look at the result, and then ask for another change on top of it, with the model remembering what came before. That is why it spread so fast among social media creators, marketers, and hobbyists. It was not just a filter. It behaved more like a cooperative editing partner.
How Nano Banana Works
Under the hood, Nano Banana is a multimodal model, which means it understands text and images together rather than treating them as separate inputs. When you give it a photo and a sentence, it builds an internal understanding of what is in the image — the subject, the scene, the style, the lighting — and then produces a new image that follows your instruction while preserving the parts you did not ask to change.
In practice, the workflow is simple:
- Provide an image (or several). You can start from a photo, an illustration, or a product shot. Nano Banana 2.1 accepts up to 14 reference images in a single request, covering up to four characters and ten objects, which is useful when you want to combine elements from different pictures.
- Describe the change in plain language. There is no special syntax to learn. “Remove the person in the background,” “make it look like a 1970s film photo,” and “put this jacket on the model” are all valid instructions.
- Review and iterate. If the result is close but not right, you refine it with another prompt. The model keeps the conversation context, so later edits build on earlier ones instead of starting over.
- Export and check. Images generated through Google’s tools carry an invisible SynthID watermark, which marks them as AI-generated even after common edits. On Google Cloud, the model also supports C2PA Content Credentials.
You do not need to write code to use it. The consumer experience lives inside Google’s apps, while developers can call the same capability through an API.
Key Features
Character Consistency
This is the headline feature. Nano Banana keeps a person, pet, or product visually consistent across different scenes and edits. Change the setting from a studio to a beach, swap an outfit, or age a character, and the face and defining features stay recognizably the same. For storytellers and brands, this is the difference between a one-off image and a usable series.
Multi-Turn Editing
Editing is conversational. You can restyle a room, then add furniture, then change the time of day, one step at a time, with each step building on the last. Earlier image tools often lost the thread after two or three edits. Nano Banana’s multi-turn ability is what makes longer creative sessions practical.
Image Merging
Nano Banana can blend multiple images into one coherent scene. Put a person from one photo and a pet from another into the same frame, or combine a product shot with a lifestyle background. The model reconciles lighting, perspective, and style so the composite looks intentional rather than pasted together.
Style Transfer and Design Mixing
You can take a pattern, texture, or art style from one image and apply it to something in another — a floral print onto a dress, a painting style onto a photograph, a material finish onto a product. This is popular with designers who want to test variations quickly without reshooting or redrawing anything.
World Knowledge
Because the model sits on top of Gemini’s broader knowledge, it can handle prompts that depend on real-world understanding: landmarks, historical styles, cultural references, and physical reasoning about how things should look. Ask for a scene “in the style of a Kyoto street at dusk” and it has something real to work from, not just a generic evening backdrop.
Thinking Levels (Nano Banana 2.1)
Version 2.1 adds adjustable thinking levels — minimal, medium, and high — which trade speed against care. Simple edits can run fast on minimal thinking, while tricky jobs like precise on-image text or complex multi-character scenes benefit from a higher setting. The default is medium, so developers who want predictable costs should set the level explicitly.
Nano Banana 2 vs Nano Banana 2.1: What Changed
If you have seen tutorials mentioning Nano Banana 2, here is the important update. Nano Banana 2 (model id gemini-3.1-flash-image) is deprecated, and Google has set 29 October 2026 as its shutdown date. After that date, calls to the old model id will stop working. Anyone building on Nano Banana 2 needs to move to gemini-nano-banana-2.1.
Version 2.1 is not just a rename. It improves text accuracy and layout for on-image words and infographics, strengthens multi-character and multi-turn consistency, and roughly halves the API cost of a typical image. It also changes the available resolutions: 2.1 generates at 1K, 2K, and 4K, but drops the small 512-pixel option that version 2 offered. If your workflow depended on 512-pixel outputs, you will need to move those requests to 1K or to the lighter Nano Banana 2 Lite variant.
One honest wrinkle: Google’s own documentation does not fully agree on what 2.1 is built on. The model card describes it as based on Gemini 3.6 Flash, while the API documentation describes it as an update to Gemini 3.1 Flash Image. For everyday use this does not matter, but it is worth knowing that the underlying-model naming is currently inconsistent in Google’s materials.
Where to Access Nano Banana
For most people, the easiest route is the Gemini app. Open Gemini, attach an image, and describe your edit in the chat. Image editing is part of the standard Gemini experience, with no separate app to install.
Developers and teams have several more doors:
- Google AI Studio — the browser-based workspace for trying the model and generating an API key.
- Gemini API — direct API access using the model id gemini-nano-banana-2.1.
- Vertex AI — the enterprise route on Google Cloud for production workloads.
- Google Flow, Stitch, and Google Ads — Nano Banana 2.1 is also rolling out across Google’s creative and advertising tools, so you may meet it inside products you already use.
- AI Mode in Google Search — the model is rolling out there as well.
If you are following an older tutorial that names gemini-3.1-flash-image, treat it as out of date. New work should target gemini-nano-banana-2.1 from day one.
Pricing and Usage Limits
Pricing depends on which door you use, and the two worlds — consumer app and developer API — are billed differently.
Developer API (Nano Banana 2.1). The API charges by the token: about $1.50 per million input tokens and $30 per million output tokens. In practical terms, that works out to roughly $0.0336 for a 1K image, about $0.0504 at 2K, and about $0.0756 at 4K. That is roughly half what Nano Banana 2 cost for the same output, which makes 2.1 the strongest quality-per-dollar option in Google’s image lineup. Batch API pricing can cut those figures further. Note that input-heavy calls — especially ones with many reference images — cost more, because you pay for the input tokens too.
Consumer app (Gemini). The Gemini app includes image editing in its regular plans rather than charging per image. At the original 2025 launch, free users were reported to get around 100 image edits per day and paid users up to 1,000, but current limits for the 2.1 era have not been confirmed in the sources used for this guide, and Google adjusts these caps over time. Check the current Gemini plan page before promising yourself a specific daily allowance.
The original model, for comparison. The first Nano Banana release (Gemini 2.5 Flash Image) was priced at $30 per million output tokens, with each image counting as 1,290 output tokens — about $0.039 per image. That figure is historical now; 2.1’s per-image costs above are the ones that matter for new work.
What Nano Banana Is Good For
The tool shines anywhere you need the same subject more than once. Social media creators use it to keep a consistent character or mascot across posts. E-commerce sellers test product shots in different settings without a photo shoot. Authors and educators build illustrated stories where the hero looks the same on every page. Marketers generate campaign variations — same product, different season, different audience — in an afternoon instead of a week. Real estate and interior designers restyle rooms to show clients options. And casual users mostly do what everyone does first: put themselves in a movie poster, change a photo’s season, or see their pet in a royal portrait.
It is less suited to jobs that need pixel-perfect control, like retouching a photo for print at exact specifications, or to anything where an AI-generated look is unacceptable. The output is impressive, but it is still AI output, and close inspection usually shows it.
Limitations to Know Before You Rely On It
- Watermarking. Every image carries an invisible SynthID watermark. If your workflow requires unmarked output, this is a hard constraint, not a setting you can switch off.
- Resolution changes. Version 2.1 drops 512-pixel output. Workflows built around tiny images need adjusting.
- People and grounding. The model’s web-grounding feature cannot use real-world photos of people from the web, which limits some research-style workflows.
- Not a photo editor. It cannot match the precise, layer-based control of Photoshop for professional retouching. Think of it as a creative generator with editing skills, not a replacement for manual editing.
- Shifting limits. Consumer caps, rollout availability, and API details change as Google iterates. Anything in this guide dated late October 2026 should be re-checked if you are reading it months later.
Frequently Asked Questions
What is Nano Banana in simple terms?
Nano Banana is Google’s AI image editing model family. You give it a photo and a plain-language instruction, and it edits the photo while keeping the subject looking like itself. It started as a secret codename and is now the everyday name for the models behind image editing in the Gemini app.
Is Nano Banana the same as Gemini 2.5 Flash Image?
Originally, yes. “Nano Banana” was the codename, and Gemini 2.5 Flash Image was the official release name. The family has since moved on: Nano Banana 2 followed, and the current version is Nano Banana 2.1 (model id gemini-nano-banana-2.1). When people say “Nano Banana” today, they usually mean the current 2.1 model.
Is Nano Banana free to use?
In the Gemini app, image editing is included in the regular Gemini experience, with daily limits that depend on your plan. Developers using the API pay per image — roughly $0.0336 for a 1K image on Nano Banana 2.1. There is no separate “Nano Banana” subscription to buy; you access it through Gemini or Google’s developer platforms.
What is the difference between Nano Banana 2 and Nano Banana 2.1?
Version 2.1 is the successor: better text accuracy, stronger multi-character consistency, roughly half the API price, and 1K/2K/4K output. Version 2 is deprecated and shuts down on 29 October 2026, so new work should use 2.1. The main thing 2.1 loses is 512-pixel output.
Can Nano Banana keep my face the same across many images?
That is its signature strength. Character consistency is the feature Google built the model around, and version 2.1 improves it further, especially across multiple turns and multiple characters. Results are strong but not perfect — hands, small text, and unusual angles can still go wrong, so check each output before publishing.
Summary
Nano Banana went from a leaked codename to one of the most useful image tools available, and the version story matters more than the name. The original Gemini 2.5 Flash Image proved the idea, Nano Banana 2 refined it, and Nano Banana 2.1 — released 6 October 2026 — is the current, cheaper, more accurate model. With Nano Banana 2 shutting down on 29 October 2026, there is no reason to start anything on the old version.
For casual creators, the Gemini app is the front door: upload a photo, describe the edit, iterate. For developers and teams, the API offers the same capability at roughly three cents per 1K image, with reference-image support that makes consistent series practical. Keep the watermark and the shifting consumer limits in mind, and Nano Banana 2.1 is one of the best value image tools you can use today.
