By Harry | Published: November 13, 2026
Nano Banana Pro is Google’s premium image generation and editing model, officially called Gemini 3 Pro Image (model ID gemini-3-pro-image). It launched in November 2025, became generally available in May 2026, and as of November 2026 it’s still the top-tier “Pro” image model in the Gemini family, even though newer and cheaper Flash models now sit beside it.
This guide covers what the model actually is, what “GemPix 2” refers to, where you can use it, what it costs, how its watermarks work, and whether something newer has replaced it.
Key Takeaways
- Nano Banana Pro and Gemini 3 Pro Image are the same model. Google uses the nickname in consumer products and the formal name in developer docs.
- “GemPix 2” was a leaked internal name reported in November 2025. Google never used it for a product, and sites selling “GemPix 2” access are third parties.
- In the Gemini API, a Nano Banana Pro image costs $0.134 at 1K or 2K and $0.24 at 4K, with no free API tier. Batch mode halves those rates.
- Every image carries an invisible SynthID watermark. At launch, Google said free and Google AI Pro users in the Gemini app also get a visible sparkle mark, while Ultra subscribers and AI Studio users don’t.
- No Pro successor exists yet. Nano Banana 2.1 (October 2026) is newer and cheaper, but Google still positions Pro as its model for complex, accuracy-critical work.
Nano Banana Pro Is Gemini 3 Pro Image
Google announced Nano Banana Pro on November 20, 2025, describing it as its state-of-the-art image generation and editing model, built on Gemini 3 Pro. That last part is the key detail. Instead of a separate image engine, it’s a Gemini model that reasons about your prompt and then produces the image itself.
So if you searched for “Gemini Banana Nano Pro,” you’re looking for the same thing. The word order varies a lot online, but there’s only one model behind it. Here’s how its official status has moved:
| Date | What happened |
|---|---|
| November 20, 2025 | Launched as a paid preview (gemini-3-pro-image-preview) in the Gemini app, Gemini API, AI Studio, and Vertex AI |
| February 26, 2026 | Nano Banana 2 replaces Pro as the default image model in the Gemini app; paid subscribers keep Pro access |
| May 28, 2026 | Stable gemini-3-pro-image becomes generally available |
| June 25, 2026 | The preview model ID is shut down in the Gemini API |
| November 2026 | Still listed as stable, with no shutdown date announced |
Those dates come from Google’s Gemini API release notes and deprecations page. If you have old code that still calls the preview ID, it won’t work anymore. Switch it to gemini-3-pro-image.
GemPix 2: A Leaked Codename, Not a Product
“GemPix 2” (often written “gempix2”) started circulating in early November 2025. TechRadar reported on November 11, 2025 that a preview tied to the Gemini app had surfaced on X, and that the model was “internally known as GEMPIX 2.” Testers also found traces of it in Whisk Labs, a Google experiment.
Nine days later, Google launched Nano Banana Pro. In February 2026, it then released a model actually named Nano Banana 2. Google has never said which shipped model the GemPix 2 strings belonged to, and it has never used the name in any announcement, doc, or pricing page we checked.
That matters for one practical reason. Search “gempix2” today and you’ll find sites using the name as a brand, often with credit packs and sign-up offers. None of them are Google. Some may pass your prompts to Google’s API, and some may not. If you’re considering one, read our explainer on how multi-model AI platforms work before you pay, because you can usually reach the real model directly through Google.
What the Model Can Do
Google’s developer announcement and the enterprise model page list the main capabilities. These are the company’s descriptions, not independent test results.
- Output resolution: 1K, 2K, and 4K.
- Aspect ratios: 14 options, from 1:1 and 16:9 to wide formats like 21:9 and extreme ones like 8:1 and 1:8.
- Reference images: up to 14 per prompt. Google’s image docs break that down as up to 6 object images, 5 character images, and 3 style references.
- Text in images: Google says it renders legible text and can translate text on menus, signs, or packaging while keeping the original layout.
- Search grounding: it can pull current information from Google Search, which helps with things like infographics or weather charts.
- Creative controls: prompts can direct lighting, camera angle, focus, color grading, and composition.
The reference limit is the feature most people underuse. Say you run a small ecommerce brand and need the same product shot in five seasonal settings. You can feed in the product photo, a logo, and a style reference in one request instead of fighting a text prompt. Our piece on AI image fusion explains the different ways tools combine multiple images, since “blend” means different things in different products.
One technical detail surprises developers. According to Google’s image generation docs, Gemini 3 image models run a thinking step that can’t be turned off in the API. It may produce up to two interim “thought” images before the final one, and thinking tokens are billed. The model’s knowledge cutoff is January 2025, so search grounding is what keeps time-sensitive visuals accurate.
What Changed From the Original Nano Banana
The original Nano Banana is Gemini 2.5 Flash Image (gemini-2.5-flash-image), the fast, low-cost model that went viral in 2025. Pro was never a minor update to it. It moved the image model onto a larger reasoning model and traded speed and price for precision.
| Feature | Nano Banana (Gemini 2.5 Flash Image) | Nano Banana Pro (Gemini 3 Pro Image) |
|---|---|---|
| Base model | Gemini 2.5 Flash | Gemini 3 Pro |
| API price per image | $0.039 | $0.134 (1K/2K), $0.24 (4K) |
| 1K/2K/4K size options | Not listed in current docs | Yes |
| Search grounding | No | Yes |
| Thinking step | No | Yes, always on |
| Status (Nov 2026) | Legacy; earliest Gemini API shutdown March 15, 2027 | Stable, no shutdown date |
The price gap is real: Pro costs roughly 3.4 times as much per 1K image. Google itself says the original model is the faster, cheaper option, and Pro is for work where accuracy matters more than speed. And if you mostly make quick variations of one image, read about why seeds don’t control small image changes before paying Pro prices for each retry.
Where You Can Use It
Gemini app
Since February 2026, the Gemini app makes its images with the Nano Banana 2 family by default. Google’s Gemini Apps limits page lists “Redo images with Nano Banana Pro” for Google AI Plus, Pro, and Ultra plans, but not for users without a plan. In practice, you generate an image, then regenerate it with Pro from the menu. The same page doesn’t give a fixed daily image count. It only says higher plans get higher limits and that limits may change.
Google AI Studio and the Gemini API
Developers call gemini-3-pro-image through the Gemini API, and you can try prompts in Google AI Studio. There’s no free tier for this model on the Gemini API pricing page, so you’ll need billing turned on.
Gemini Enterprise Agent Platform (formerly Vertex AI)
Google Cloud rebranded Vertex AI as Gemini Enterprise Agent Platform in 2026, and that’s where enterprise teams use Nano Banana Pro. The May 28, 2026 GA announcement said 1K and 2K output are covered by the enterprise SLA, while 4K output was still in preview. The model runs in the global endpoint, and Google lists its earliest retirement as May 28, 2027 or later.
Other Google products
At launch, Google also brought Pro to AI Mode in Search for US subscribers, NotebookLM, Google Ads, Slides and Vids for Workspace customers, and Flow for Ultra subscribers. Several of those surfaces have since moved to newer Flash models as defaults, so check the specific product before assuming which model you’re getting.
Nano Banana Pro Pricing and Limits
Here are the official per-image prices as of November 2026. The Gemini API and Google Cloud list the same standard rates.
| Item | Standard | Batch |
|---|---|---|
| Output image, 1K or 2K | $0.134 | $0.067 |
| Output image, 4K | $0.24 | $0.12 |
| Text and image input | $2.00 per 1M tokens (about $0.0011 per input image) | $1.00 per 1M tokens |
| Text and thinking output | $12.00 per 1M tokens | $6.00 per 1M tokens |
Sources: Gemini API pricing and Google Cloud generative AI pricing. Note that 1K and 2K cost the same, so there’s rarely a reason to request 1K for paid work.
A quick example. A marketing team making 400 social graphics a month at 2K would pay about $53.60 in output fees at standard rates, or about $26.80 through batch if the work isn’t urgent. Thinking tokens and reference images add a little on top.
On limits, the API caps output at 32,768 tokens per request and input at 65,536 tokens, per the model page. Rate limits depend on your billing tier.
Watermarks: SynthID, the Sparkle, and C2PA
Every image made or edited with Gemini 3 Pro Image includes SynthID, Google’s invisible watermark. You can’t see it, and you can’t switch it off. Google added a checker to the Gemini app in November 2025: upload an image and ask whether Google AI made it.
The visible watermark is separate. In its launch post, Google said images from free and Google AI Pro users keep the visible Gemini sparkle, while Google AI Ultra subscribers and images created in Google AI Studio don’t get it. That policy could change, so check your own outputs.
Google is also adding C2PA Content Credentials, an industry metadata standard that records how an image was made. The enterprise model page lists C2PA as supported. And in February 2026, Google said C2PA verification was coming to the Gemini app, without giving a date.
Is There a Newer Version Than Nano Banana Pro?
Not a newer Pro. As of November 2026, Google’s model list still shows gemini-3-pro-image as the only Pro image model, and its model page lists November 2025 as the latest model update.
What has arrived is a faster tier beneath it:
- Nano Banana 2 (Gemini 3.1 Flash Image), February 26, 2026. Google described it as Pro-level capability at Flash speed. It’s now deprecated in favor of 2.1.
- Nano Banana 2 Lite, June 30, 2026. The fastest and cheapest option, at $0.034 per 1K image, with 1K output only.
- Nano Banana 2.1, generally available October 6, 2026. Google’s docs call it the primary high-efficiency model and the more efficient option next to Pro. Google Cloud lists it at about $0.034 per 1K image.
Google’s own docs still describe Pro as the premium choice for complex tasks, strong world knowledge, and brand consistency. Some Google benchmarks show the Flash models scoring close to or above Pro, but those are the company’s own tests. If you want an outside view of how Pro compares with a rival, our Nano Banana Pro vs FLUX.2 comparison covers that.
When Nano Banana Pro Is Worth Paying For
For most everyday images, start with Nano Banana 2.1. It costs about a quarter as much per 1K image, and Google recommends it for new projects. Move up to Nano Banana Pro when a mistake would be expensive: text-heavy graphics, infographics built on real data, product images that must match a brand, or scenes where up to five people need to stay recognizable.
A simple rule works well. Draft and explore on 2.1, then run the final version through Pro and compare. If you can’t see a difference for your use case, you’ve just saved yourself the extra cost. And if you’re choosing among image tools more broadly, our AI image generator comparison covers options outside Google.
FAQ
Is Nano Banana Pro free to use?
Not fully. Google’s Gemini Apps limits page lists Pro regeneration only for AI Plus, Pro, and Ultra plans, and the Gemini API has no free tier for gemini-3-pro-image. Free Gemini users get the Nano Banana 2 family instead, which Google built to bring many Pro features to its faster, cheaper tier.
Can I remove the SynthID watermark from Nano Banana Pro images?
No. SynthID is embedded in every image the model creates or edits, and Google offers no setting to turn it off. Only the visible sparkle depends on your plan. If your workflow needs AI disclosure anyway, SynthID simply backs up what you already label.
Are gempix2 websites run by Google?
No. Google never launched a product called GemPix 2, so any site branded that way is a third party. Some resell access to Google’s real models through the API, often with markup. Before paying, check whether the site names the exact Google model ID and how it bills you.
Why does my code using gemini-3-pro-image-preview return an error?
Because Google shut that preview ID down on June 25, 2026. The replacement is the stable gemini-3-pro-image model, which Google released on May 28, 2026. In most cases, changing the model string is enough, but recheck your pricing assumptions and any preview-only parameters.
Does Nano Banana Pro work in languages other than English?
Yes, for in-image text. Google’s developer announcement highlights translating text on menus, signs, and packaging while keeping the original style. Google doesn’t publish an official list of supported in-image languages, so test your specific language and check spelling before you publish anything client-facing.
Harry covers artificial intelligence for AI Journal Now.
