GoogleのNano Banana 2、Proレベルの画像生成をFlash速度で最大40%低コストのAPI価格で実現

Google is introducing Nano Banana 2, a new image generation model that pairs the capabilities of the pricier Pro model with the speed of Gemini Flash. It's set to become the default in the Gemini app.

Technically, image generation runs on the new multimodal Gemini 3.1 Flash Image model. Google says it combines the advanced capabilities of Nano Banana Pro, which shipped in November, with the speed of the Flash model series.

Google says the model draws on Gemini's knowledge base and pulls real-time information and images from web searches to render subjects more accurately. Just like Nano Banana Pro, it can also create infographics, turn notes into diagrams, and generate data visualizations.

Nano Banana 2 cuts API costs by up to 40 percent

Google says it improved instruction following so the model handles complex prompts more precisely. On the specs side, Nano Banana 2 supports multiple aspect ratios and resolutions ranging from 512 pixels up to 4K. Google also promises more vivid lighting, richer textures, and sharper details compared to the original Nano Banana.

Text rendering got an upgrade too: the model should now produce legible text for things like marketing mockups or greeting cards, including the ability to translate and localize text within an image.

These are all features the Pro model already offers. But the new model, running under the API name gemini-3.1-flash-image-preview, comes in significantly cheaper according to Google's price list:

At the highest resolution, that's a savings of nearly 40 percent, a compelling price gap for developers working through the API.

Nano Banana Pro becomes a niche tool in the Gemini app

Nano Banana 2 replaces Nano Banana Pro as the default image generation model in the Gemini app across Fast, Thinking, and Pro modes. Google AI Pro and Ultra subscribers still get access to Nano Banana Pro for specialized tasks, but they'll have to select it manually through the three-dot menu. That effectively turns the Pro model into a niche tool for users who need maximum detail.

In our initial testing, Nano Banana 2 is a clear step up from the original Nano Banana, which ran on Gemini 2.5 Flash, and it holds its own against the Pro version. It renders our complex benchmark prompt largely correctly, even nailing the trickiest detail: the horse is supposed to ride on the human instead of the other way around. The model has seen the normal scenario—humans riding horses—millions of times during training, while the reversed version has likely rarely or never appeared in its training data. Until now, only Nano Banana Pro could consistently pull that off. That said, the Pro output still looks more dynamic and realistic overall, while Nano Banana 2 has a slightly artificial quality to it.

If you're after maximum image quality, Nano Banana Pro is likely still the gold standard. Google seems to agree—otherwise, there'd be no reason to keep offering it to paying subscribers. But for everyday images and quick edits, Nano Banana 2 looks like a solid alternative at a fraction of the cost.

Beyond the Gemini app, Nano Banana 2 is rolling out across a range of Google products: Google Search via AI Mode and Lens, with 141 new countries and territories and eight additional languages coming online. Developers can try the model as a preview in AI Studio and through the Gemini API, as well as in Vertex AI on Google Cloud. In Google's creative tool Flow, Nano Banana 2 becomes the new default, available to all users at no credit cost. The model will also power campaign creation suggestions in Google Ads.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.

Subscribe now