Icon

Updated weekly

OW

OW

Icon

Updated weekly

OW

Output Weekly

Output Weekly

Tools, workflows and honest rankings for people who make things with AI

Tools, workflows and honest rankings for people who make things with AI

Images

Nano Banana explained: every model, what it does, and where to use it (November 2026)

Ravi Menon

Yellow bananas on a blue surface

Key takeaways

  • Nano Banana is Google's name for Gemini's built-in image generation. As of November 2026 the name covers five models, Nano Banana 2.1, Nano Banana 2, Nano Banana 2 Lite, Nano Banana Pro, and the original Nano Banana.

  • Nano Banana 2.1, released October 6, 2026, is the model Google now recommends for new projects. Through the Gemini API it costs $0.0336 for a 1K image, about half the price of Nano Banana 2.

  • Nano Banana 2 and 2.1 take up to 14 reference images, split as four character images and 10 object images. Nano Banana Pro takes five character images, six object images, and three style references.

  • In the Gemini app, Nano Banana 2 is free with usage limits, and Nano Banana 2.1 needs a Google AI plan. The API has no free tier for any of the image models.

  • Every image carries an invisible SynthID watermark. The Gemini app's visible watermark can be switched off by most users, but not in India, South Korea, or Vietnam without AI Ultra.

What is Nano Banana?

Nano Banana is the name Google uses for the image generation built into its Gemini models. The same model reads your text, your photos, and the conversation so far, then returns an image, which is why you can edit a picture by describing the change in plain words.

The name started with one model, Gemini 2.5 Flash Image, and now covers a family of five. Google's image generation docs list them by API model ID, which is the surest way to tell them apart, since the Nano Banana version numbers don't match the Gemini versions underneath. Nano Banana 2.1, for instance, is "based on Gemini 3.6 Flash," per its model card.

Imagen, Google's older image family, is gone from the Gemini API, so Nano Banana is now the only image generation option there.

The Nano Banana lineup at a glance

Model

API model ID

Stable release

Max output

Price per 1K image

Nano Banana 2.1

gemini-nano-banana-2.1

October 6, 2026

4K

$0.0336

Nano Banana 2 Lite

gemini-3.1-flash-lite-image

June 30, 2026

1K

$0.0336

Nano Banana 2

gemini-3.1-flash-image

May 28, 2026

4K

$0.067

Nano Banana Pro

gemini-3-pro-image

May 28, 2026

4K

$0.134

Nano Banana (original)

gemini-2.5-flash-image

October 2, 2025

1K

$0.039

Dates are stable releases from Google's deprecations page. The earlier preview IDs for Nano Banana 2 and Pro shut down on June 25, 2026. Prices are standard paid-tier rates on the Gemini API pricing page.

Nano Banana 2.1, the current default

Nano Banana 2.1 is an update to Nano Banana 2, and Google's docs say it's "Recommended for all new projects." It generates at 1K, 2K, or 4K (1K is the default), takes text, images, video, and PDFs as input, and supports multi-turn editing, which the docs describe as the recommended way to iterate on an image.

Its reference limit is 14 images. Up to four can be characters, for keeping faces consistent, and up to 10 can be objects that need to show up accurately in the result. It can also ground an image in Google Search, pulling from both web results and Google Image Search, so a prompt like a five-day weather chart for a real city gets real data. One catch from the docs is that Search grounding on this model doesn't use real-world images of people from the web.

Its thinking level can be set to minimal, medium, or high, with medium as the default.

The model card also names what still goes wrong. Small text is "often blurry in 1k model," long paragraphs render poorly, and character consistency "is not always perfect between input images and generated output image." It also lists occasional left-right mix-ups, partial instruction following in doodle-based edits, and limited 3D reasoning. For posters with small type, render above 1K.

Nano Banana 2 and Nano Banana 2 Lite

Nano Banana 2 (Gemini 3.1 Flash Image) was the workhorse model before 2.1 arrived. It has the same 14-image reference limit, the same split of four characters and 10 objects, the same Search and Image Search grounding, and the same extra-wide aspect ratios (1:4, 4:1, 1:8, and 8:1). It's the only Nano Banana model that also renders a 512px image, which costs $0.045.

The odd part is its price. At $0.067 for a 1K image and $0.151 at 4K, Nano Banana 2 costs about double what its replacement does.

Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) is built for speed and volume. Google says it targets "sub-2 second" latency, and it outputs 1K images only, at $0.0336 each, or half that through the Batch API. The docs say it's "Not optimized for multiple reference inputs or multi-turn sequential editing," and it doesn't support Search grounding. Google pitches it for fast edits like color swaps, stickers, and background changes.

Nano Banana Pro

Nano Banana Pro (Gemini 3 Pro Image) is the premium model, aimed at complex instructions, brand consistency, and accurate text. Like the other Gemini 3 image models, it reasons through the prompt before drawing, a step Google says "cannot be disabled in the API." Along the way it can produce up to two interim images to test the composition.

Pro handles references differently from the Flash models. It takes up to six object images, up to five character images, and up to three style reference images, and it's the only Nano Banana model with style references. It supports Search grounding on web results, but not Image Search, and the extra-wide aspect ratios aren't documented for it.

Pro costs $0.134 for a 1K or 2K image and $0.24 at 4K. There's no Batch pricing for it. Google's Nano Banana 2.1 model card puts the thinking version of 2.1 ahead of Pro on every text-to-image and editing score it reports, including multi-character consistency. What Pro still has that 2.1 doesn't is style references.

The original Nano Banana

The model that started the name, Gemini 2.5 Flash Image, is now listed as legacy. It's deprecated and shuts down on March 15, 2027, and Google points its users to Nano Banana 2 Lite as the migration path. It costs $0.039 per image up to 1024x1024 pixels, and the docs say it "works best with up to 3 images as input."

What each Nano Banana model costs in the API

Model

512px

1K

2K

4K

Batch, 1K

Nano Banana 2.1

Not offered

$0.0336

$0.0504

$0.113

$0.0168

Nano Banana 2 Lite

Not offered

$0.0336

Not offered

Not offered

$0.0168

Nano Banana 2

$0.045

$0.067

$0.101

$0.151

Not listed

Nano Banana Pro

Not offered

$0.134

$0.134

$0.24

Not listed

Nano Banana (original)

Not offered

$0.039

Not offered

Not offered

$0.0195

These are output costs per image. Input images and prompt text are billed separately as tokens, and thinking tokens are billed too. Search grounding is free for the first 5,000 requests a month, shared across all Gemini 3 and newer models, then $14 per 1,000 requests.

There's no free API tier for any Nano Banana model. Every one lists its free-tier price as "Not available." Google's pricing page does say that using Google AI Studio itself is free of charge in supported regions.

A 4K image from Nano Banana 2.1 costs less than a 2K image from Pro. Our FLUX 3 explainer has per-image API prices for another current model family if you're comparing.

Where you can use Nano Banana

The Gemini app is where most people meet Nano Banana. Per Google's Gemini image generation page, Nano Banana 2 is "Available for free," and any Google AI plan unlocks Nano Banana 2.1, which Google calls its best image model. Google AI Plus, Pro, and Ultra subscribers can also regenerate an image with Nano Banana Pro through a "Redo with Pro" option in the three-dot menu. The app page lists 2K resolution for both models, so 4K output is an API feature.

Google doesn't publish a daily image count for the app. Its help page describes compute-based limits that refresh every five hours up to a weekly cap. Plus gets twice the standard limit, Pro four times, and Ultra five or 20 times the Pro limit depending on the subscription. Our ranking of mobile AI image generation apps covers the Gemini app on iPhone and Android next to six others.

For developers, Nano Banana runs in Google AI Studio and the Gemini API. On Google Cloud, Vertex AI is now part of Gemini Enterprise Agent Platform, and all five models are available there. Nano Banana 2.1 went generally available on that platform on October 6, 2026, with up to 14 images per prompt, per its Agent Platform model page.

The 2.1 model card also lists Google Search AI Mode, Google Ads, Google Flow, and Google Stitch as places the model runs, and third-party apps and API hosts carry Nano Banana models at their own prices.

How to prompt Nano Banana

Nano Banana models don't take a negative prompt. Google's guide has a tip titled "Use 'Semantic Negative Prompts'," which means describing the scene you want in place of listing what to leave out. Instead of "no cars," it suggests "an empty, deserted street with no signs of traffic." Our explainer on whether negative prompts still matter covers why most hosted APIs dropped the field.

For text inside an image, Google's docs say the model works best if you generate the text first and then ask for an image that includes it. For several images, the docs warn that the model "won't always follow the exact number of image outputs" you request, so ask for one at a time when the count is fixed.

For a recurring character, give each reference a single job and stay inside the character limits, which are four images on 2.1 and 2 and five on Pro. The Gemini API doesn't save a character between requests, so you send the references each time. Our ranking of AI image generators for character consistency compares Nano Banana with tools that do save one.

Watermarks, SynthID, and the Imagen shutdown

Every image from every Nano Banana model carries an invisible SynthID watermark, according to Google's docs. Images made in the Gemini app also get Content Credentials (C2PA) metadata.

The visible watermark is a separate setting. Google's watermark help page puts the switch under Settings, then Media Watermark, and it covers images, video, and music made in the Gemini app. In India, South Korea, and Vietnam, the setting only appears for AI Ultra subscribers, and everyone else there gets visible watermarks automatically. Turning it off doesn't touch SynthID or Content Credentials. Flow has its own "Visible watermarking" toggle with a similar rule for those three countries.

The docs now call Imagen "shut down and no longer available through the Gemini API," and the deprecations page shows the last Imagen 4 models switched off on August 17, 2026, with Nano Banana 2.1 named as the replacement.

Nano Banana FAQ

What is the newest Nano Banana model?

Nano Banana 2.1, model ID gemini-nano-banana-2.1, released October 6, 2026. It's an update to Nano Banana 2 and is the model Google recommends for new projects in the Gemini API.

Is Nano Banana free?

In the Gemini app, Nano Banana 2 is free with usage limits, and Nano Banana 2.1 needs a Google AI plan. In the Gemini API, there's no free tier for any Nano Banana model, and Nano Banana 2.1 starts at $0.0336 per 1K image.

How many reference images can Nano Banana use?

Nano Banana 2.1 and Nano Banana 2 take up to 14, with up to four images of characters and up to 10 of objects. Nano Banana Pro takes up to five character images, six object images, and three style references. Nano Banana 2 Lite isn't built for multiple references.

Can Nano Banana make 4K images?

Yes, through the API on Nano Banana 2.1, Nano Banana 2, and Nano Banana Pro. Nano Banana 2 Lite and the original model stop at 1K, and the Gemini app lists 2K as its resolution.

Can you remove the Nano Banana watermark?

The visible watermark, yes, in the Gemini app's Media Watermark setting, except in India, South Korea, and Vietnam without AI Ultra. The invisible SynthID watermark stays on every image regardless.