Google Introduces Nano Banana 2 Image Generation Model with Enhanced 4K Output and Reduced Cost

Google has released Nano Banana 2, an image generation model that delivers 4K ultra-high-definition quality at high speeds, while halving the operational cost compared to its predecessor. The model, powered by the new Gemini 3.1 Flash Image, has quickly gained attention across the AI community for its capabilities.

Hands sketching a game UI on a tablet, demonstrating rapid image generation and design capabilities.

Nano Banana 2 is designed for rapid image generation, features improved multilingual text processing, and can integrate real-time internet information to produce high-resolution visuals. Initial tests and evaluations indicate its strong performance in various benchmarks.

Enhanced Performance and Cost Efficiency

The model has demonstrated the ability to generate complex visuals, such as game user interfaces from simple text prompts, and convert sketches into web UI designs. It can also produce multiple comic strips efficiently. A notable improvement is its accurate rendering of non-English characters, eliminating common text distortions seen in previous AI image generators.

Infographic showing Nano Banana 2 leading in text-to-image generation benchmarks.

In benchmark tests, Nano Banana 2 achieved the top position globally in text-to-image generation, according to Artificial Analysis. For image editing, it ranked third, behind GPT Image 1.5 and Nano Banana Pro. On Image Arena, Nano Banana 2 led the text-to-image charts with 1279 Elo points, and its image editing capabilities were second only to GPT Image.

Google's internal evaluations, which included enabling "Thinking + Text Search + Image Search," showed Nano Banana 2 outperforming competitors like GPT-Image 1.5 and Grok Imagine Image Pro in overall user preference, visual quality, and infographic accuracy. It also surpassed Nano Banana Pro, its more advanced counterpart, while offering a significantly lower output price of $0.0672 per image, half the cost of Nano Banana Pro.

Key Features and Applications

Nano Banana 2 integrates Gemini's knowledge base and real-time web search capabilities. This allows it to access and interpret real-world visual references when generating images, such as architectural structures, ensuring accuracy in its renderings. This feature also supports the creation of infographics, data visualizations, and diagrams from textual notes.

Digital infographic showing a clear, top-down view of the water cycle with guiding arrows.

For example, when prompted to create a scientific infographic on the water cycle, the model produced a clear, top-down aerial view illustrating each stage of the cycle with guiding arrows and soft, even lighting. Similarly, for a comparison chart of cloud types, it generated a triptych featuring cumulus, stratus, and cirrus clouds with distinct backgrounds and prominent labels in a high-contrast comic book style.

The model can also re-interpret real-world locations, such as the Château du Clos Lucé, into specific artistic styles like Synthetic Cubism, accurately maintaining structural features while applying the requested aesthetic. Google has developed an application called "Window Seat" to showcase this capability, generating realistic airplane window views based on global scenes and real-time weather data.

Greeting card with perfectly rendered, clear text in multiple languages, demonstrating AI text accuracy.

Another significant improvement is in text rendering and translation. Nano Banana 2 addresses the issue of distorted text in AI-generated images, producing clear and accurate text suitable for marketing materials and greeting cards. It can also localize content, translating text into different languages and adjusting visual elements to match the target market, as demonstrated by the "Global Ad Localizer" tool.

Quality and Customization

Nano Banana 2 maintains high image quality despite its increased speed. It demonstrates improved subject consistency, capable of retaining consistent features for up to five characters and high fidelity for 14 objects within a single workflow. This allows for complex scenes with multiple distinct elements without visual inconsistencies. For narrative sequences, the model can maintain character appearance across multiple images while varying expressions and perspectives. The "Pet Passport" application further illustrates this, allowing users to place their pets in various global landmarks while maintaining consistent pet features.

Three images showing the same golden retriever consistently placed in different global landmarks.

The model also offers more precise instruction following, accurately capturing subtle details from complex descriptions. Its output specifications range from 512px to 4K resolution, with support for various aspect ratios, including extreme ones like 4:1 and 1:8, suitable for banners and vertical images. A new 512px resolution is optimized for low-latency, high-load scenarios.

Developers can adjust the "Thinking Level" to control the depth of reasoning the model performs before generation. While the default prioritizes speed, advanced or dynamic modes enable more thorough reasoning for complex prompts, enhancing output quality and adherence to instructions. The visual quality itself has seen improvements, with more vivid lighting, richer textures, and sharper details, as evidenced by generated images of misty valleys and Pop Art fashion portraits.

Widespread Adoption and Availability

Since its initial release, Nano Banana has been used to generate over a billion images. Nano Banana 2 has been integrated across Google's product ecosystem. It will replace Nano Banana Pro in the Fast, Thinking, and Pro models within the Gemini App, though users can still access Nano Banana Pro for specific needs.

Google has implemented daily limits for image generation: up to 100 images for non-subscribers and 1,000 images for those subscribed to Google AI Plus, Pro, or Ultra.

The model is also incorporated into Google Search, including the Google App, AI Mode on mobile and desktop, and Lens. Developers can access a preview version through AI Studio, Gemini API, and Vertex AI, with support from Google Antigravity. Flow users will find Nano Banana 2 as the default model, available for free. Additionally, it is integrated into Google Ads to provide smart suggestions for campaign creation.

Screenshot of Google AI Studio interface for developers accessing Nano Banana 2.

Google's strategy with Nano Banana 2 aims to address the daily needs of most users by offering a fast, accurate, and visually appealing image generation tool. Nano Banana Pro remains available for professional applications requiring extremely high factual accuracy, providing users with options tailored to their specific requirements.