Nano Banana 2 is Here - Faster and Cheaper

Nano Banana 2: A Detailed Overview

Key Concepts:

• Nano Banana 2: The latest image generation model from Google, built on Gemini 3.1 Flash.
• Gemini 3.1 Flash: A faster, more cost-effective version of the Gemini 3.1 model series, optimized for image generation.
• Nano Banana Pro: The previous generation image generation model, based on Gemini 3 Pro.
• Chain of Thought: The reasoning process the model uses to arrive at an image, often involving intermediate image generation steps.
• Spatial Understanding: The model’s ability to accurately arrange objects in a scene based on instructions.
• Text Rendering: The model’s capability to accurately and legibly incorporate text into generated images.
• Tool Use: The model’s ability to leverage external tools like Google Search to gather information for image generation.

Introduction & Core Capabilities

Google has released Nano Banana 2, a new image generation model representing an evolution from the popular Nano Banana Pro. The key distinction is that Nano Banana 2 is built upon Gemini 3.1 Flash, marking the first publicly available image generation model utilizing this specific iteration of the Gemini family. While its image generation capabilities are comparable to Nano Banana Pro, Nano Banana 2 offers a significant advantage: reduced operational cost. It also introduces support for new aspect ratios, expanding the range of image formats it can produce. Early access provided to the model indicates impressive performance given its cost-effectiveness.

Comparative Analysis: Nano Banana 2 vs. Nano Banana Pro

A direct comparison was conducted to assess the performance of Nano Banana 2 against Nano Banana Pro. The presenter emphasized that, due to the nature of large language models (LLMs), identical prompts will yield varying results each time. However, the overall quality remains consistently similar between the two models, particularly for straightforward prompts.

1. World Knowledge & Historical Accuracy

A prompt requesting a photorealistic image of the Eiffel Tower on its inauguration date (noon, original date) was used to test the model’s world knowledge. Both Nano Banana 2 and Nano Banana Pro successfully identified the location and generated images reflecting the historical period, including accurate attire. Nano Banana 2, like its predecessor, employs a chain of thought process, sometimes generating intermediate images during its reasoning. The image generation took approximately 19 seconds. Nano Banana Pro also produced a comparable result. Both models demonstrated the ability to accept follow-up questions; when asked to label historical buildings in the generated image, both accurately identified the Eiffel Tower, the original ballet, and a temporary exposition archway.

2. Spatial Understanding & Counting

The models were challenged with a prompt requiring the arrangement of seven macarons in a perfect circle, each a different color (red, orange, yellow, green, blue, indigo, violet) in clockwise order, with a card in the center displaying "Seven Wonders" in gold foil lettering. Nano Banana 2 and Nano Banana Pro both successfully fulfilled the instructions, generating images with seven macarons arranged correctly.

3. Text Rendering Capabilities

A significant strength highlighted for Nano Banana 2 is its superior text rendering ability. A prompt requesting a film festival poster in English, Japanese, and Arabic, each using an elegant script, resulted in a visually appealing and accurately aligned output from Nano Banana 2. The orientation of the text in each language was perfect.

Leveraging External Tools & Data Integration

Nano Banana 2’s ability to utilize external tools, specifically Google Search, was demonstrated. A prompt requesting a bar chart displaying the market capitalization of the top five most valuable companies (Nvidia, Apple, Microsoft, Alphabet, Amazon) initially produced an incorrect ordering (Nvidia, Apple, Alphabet, Microsoft, Amazon). However, a subsequent run of the same prompt yielded the correct order (Nvidia, Apple, Microsoft, Alphabet, Amazon), suggesting the results are influenced by the current Google Search findings rather than inherent model knowledge.

Technical Diagram Generation

The model was tasked with creating a technical architectural diagram illustrating how Anthropic’s Cloud API works, including message flow, system prompt injection, tool use loop, and safety filtering. Nano Banana 2 successfully generated a diagram based on documentation from claude.ai and other sources. Notably, the model avoided the common issue of distorted or illegible text often found in images generated by other LLMs. While the diagram’s complete technical accuracy requires expert review, it appears to be a correct rendering of a technical diagram.

Complex Prompts & Creative Applications

Recipe Card Generation

A complex prompt requesting a professional recipe card for a classic French croissant, including seven ingredients with precise measurements and six numbered steps with nutritional information, was used. Both Nano Banana 2 and Nano Banana Pro generated comparable and accurate recipe cards.

Comic Strip Creation

Nano Banana 2 was tested with a four-panel comic strip prompt. The model demonstrated strong character consistency and accurate text placement. A minor issue was observed where the text was duplicated in one panel. Nano Banana Pro also produced a satisfactory result, though Nano Banana 2 was noted to render the coffee shop setting slightly better.

Unrealistic Scenario: Underwater Piano

A highly imaginative prompt requesting a hyperrealistic photograph of a transparent glass grand piano in a coral reef, played by an octopus with correctly positioned tentacles, and surrounded by tropical fish, was presented. Nano Banana 2 successfully generated an image fulfilling all requirements, including the piano’s branding and realistic underwater lighting. The anatomical correctness of the octopus’s tentacles (including reflections suggesting additional tentacles) was also noted.

Pricing & Conclusion

The primary advantage of Nano Banana 2 is its anticipated lower cost compared to Nano Banana Pro, while maintaining comparable image quality. Specific pricing details were not disclosed due to potential changes, but a link to the official pricing page was promised in the video description. The presenter concluded that Nano Banana 2 offers a strong performance-to-cost ratio and encouraged viewers to experiment with the new model. As stated by the presenter, “I think you will notice really good performance to cost ratio for this model.”