Nano Banana 2: What Google’s Image Model Actually Costs and Where You Can Use It
Google’s image generation model Nano Banana 2 has moved from a February preview to a fully supported product, and the details that matter to businesses resolutions, per-image prices and enterprise availability are now published rather than estimated.
The model is the successor to the original Nano Banana, the image tool that spread widely across social platforms in 2025. Its formal name is Gemini 3.1 Flash Image. Google announced it on 26 February 2026, positioning it as a model that keeps the quality of the higher-end Nano Banana Pro while running at the speed of Google’s Flash tier.
The rollout timeline
Three dates define the current state of the product.
On 26 February 2026, Google released the model into the Gemini app, Google AI Studio, the Gemini API in preview, Vertex AI in preview, Flow and Google Ads. Google also said image generation in AI Mode and Lens expanded to 141 new countries and territories and eight additional languages.
On 28 May 2026, Google Cloud confirmed that Nano Banana 2 and Nano Banana Pro were generally available on its enterprise platform. General availability covers 1K and 2K output. 4K output remains in preview status, which is an important distinction for teams that need a service-level agreement.
On 30 June 2026, Google added a cheaper sibling, Nano Banana 2 Lite, alongside a separate video model called Gemini Omni Flash.
What the model does
Nano Banana 2 generates images from text and edits images that a user uploads. Google’s developer documentation lists four output sizes: 512px, 1K, 2K and 4K. The 512px tier is specific to Nano Banana 2 and is aimed at cheap, fast drafts. Google also added unusually wide and tall aspect ratios — 4:1, 1:4, 8:1 and 1:8 — which suit banners, headers and vertical mobile layouts.
Two features separate it from most rival image tools.
The first is grounding. Nano Banana 2 can pull from Google Search and Google Image Search while generating, which helps when a prompt refers to a real place, product or public event. The cheaper Lite version does not support this.
The second is a thinking control. Developers can set the model to “minimal” for speed or “high” for harder prompts. Minimal is the default.
For consistency work, Google’s API documentation states that Nano Banana 2 accepts up to 10 object reference images and up to 4 character references in a single request. Nano Banana Pro takes up to 6 objects, 5 characters and 3 style references. Nano Banana 2 Lite takes up to 14 object images but only outputs at 1K.
A preview feature added in May lets Nano Banana 2 take a video file as an input and produce a still image from it, which Google frames as a way to build thumbnails and infographics from existing footage.
Official pricing
Google prices image output by token, then publishes an effective per-image figure. Based on the current Gemini API pricing page, the standard rates for Nano Banana 2 are:
| Output size | Standard price per image | Batch price per image |
|---|---|---|
| 512px (0.5K) | $0.045 | $0.022 |
| 1K | $0.067 | $0.034 |
| 2K | $0.101 | $0.050 |
| 4K | $0.151 | $0.076 |
Batch processing runs at half the standard rate. There is no free tier for these image models on the paid API.
For comparison, Nano Banana Pro costs $0.134 per image at 1K or 2K and $0.24 at 4K. Nano Banana 2 Lite costs about $0.034 per image at 1K, its only supported size.
Those numbers change the maths for high-volume users. A team producing 10,000 product images a month at 1K would pay roughly $670 on the standard rate, or about $340 using batch requests.
Consumers do not pay per image. Image generation sits inside the Gemini app under a “Create images” tool, with Fast, Thinking and Pro options. Google AI Pro and Ultra subscribers can regenerate a result using Nano Banana Pro from the three-dot menu. Google has not published a public per-day image cap for free accounts, so anyone planning around free usage should treat limits as subject to change.
A number of independent web tools also run on the model through Google’s API. Browser-based services such as Nano Banana 2 package it with credit-based plans, editing presets and export options for users who do not want to work with an API key. These are third-party products and are not operated by Google, so buyers should check credit costs and commercial-use terms before subscribing.
Provenance and the honest drawbacks
Every image Google generates carries an invisible SynthID watermark plus a visible marker. Google says its SynthID verification tool has been used more than 20 million times since launching in November, and that C2PA Content Credentials support is being added to the Gemini app.
The drawbacks are worth stating plainly. 4K is still a preview capability rather than a supported one, so it is not the right choice for a production pipeline that needs guaranteed uptime. Grounding raises quality on real-world subjects but adds latency. And the visible watermark on consumer output is a genuine limitation for anyone who needs clean assets that generally means moving to the paid API or an enterprise deployment.
There is also a naming problem that affects buyers directly. “Nano Banana 2”, “Nano Banana Pro” and “Nano Banana 2 Lite” are three different models at three different prices, and Pro is not simply a better version of 2. Pro is built on Gemini 3 Pro Image, while Nano Banana 2 is built on Gemini 3.1 Flash Image. Teams comparing quotes from agencies or software vendors should ask which model ID is actually being called.
What it signals
The direction is cost, not novelty. A 512px tier at $0.045 and a Lite model at roughly $0.034 per image suggest Google expects image generation to become a background utility inside other software rather than a destination product. For publishers, retailers and marketing teams, the practical question has shifted from whether the output is good enough to how many images a budget now buys.