Google Drops Nano Banana 2.1: The AI Image Generation Giant Gets Faster, Sharper, and Half the Price

Google has officially released Nano Banana 2.1, marking the latest evolution of its flagship artificial intelligence model designed to convert typed prompts into rich images or precisely edit pre-existing photographs. Announced on Tuesday, the new model introduces a suite of performance upgrades and a dramatic slashing of developer pricing as Google continues to aggressively push its creative AI tooling across its entire ecosystem of consumer applications and developer platforms.

The rollout is expansive, hitting users across the Gemini app, Google Search’s dedicated AI Mode, and Google Ads. At the same time, developers are gaining access to the upgraded model through a variety of core technical environments, including Google AI Studio, Flow, and Stitch. This wide-reaching deployment follows a trajectory that has turned Google’s image generation technology into a massive commercial and cultural touchstone over the past year.

The phenomenon originally began with the debut of the first-generation Nano Banana model, which transformed the Gemini app into a viral sensation. In September 2025, a specific viral feature that turned casual user selfies into collectible figurines catapulted Gemini to the number one spot on both major mobile app stores, temporarily dethroning ChatGPT after nearly three years of dominant supremacy at the top of the charts. That surge in popularity and utility coincided with a historic financial milestone for parent company Alphabet, whose market capitalization crossed the $3 trillion threshold during the same stretch.

Following that success, Google released Version 2 in February, which was built upon the Gemini 3.1 Flash Image architecture. That iteration introduced a critical capability: the model was engineered to cross-reference Google Search before rendering real-world objects, entities, or historical events, substantially increasing the accuracy of generated images depicting real people or factual scenarios.

Now, Nano Banana 2.1 builds directly upon those foundations with a focus on visual fidelity, editing precision, and user control. In an official post shared on social media platform X, Google stated that the upgraded version outperforms its predecessor models across the board. The company highlighted notable leaps in visual design, mask-based editing, and subject consistency, which are designed to help creators produce significantly more natural-looking imagery.

To quantify these improvements, Google pointed to ELO preference ratings derived from blind text-to-image tests. In these evaluations, Nano Banana 2.1 secured an impressive score of 1,050 ELO points. For comparison, the previous Nano Banana 2 model achieved 990 points, while the specialized Nano Banana Pro variant registered 935 points. Within the ELO methodology, higher numbers denote superior performance in qualitative human preference evaluations without a fixed upper ceiling.

Technical specifications for Nano Banana 2.1 show substantial expansion in handling complex creative inputs. On paper, the model is capable of processing up to 14 reference images simultaneously in a single prompt session. It can track up to four distinct characters and ten individual objects across those reference assets, maintaining narrative and visual cohesion. When it comes to output, the system supports resolutions up to 4K—approaching standard television sharpness—and can generate images in expansive aspect ratios as wide as 8:1.

Furthermore, developers are being given unprecedented operational control over the generation process. Creators can configure how long the model "thinks" before it begins drawing, scaling the processing time from minimal latency to deep, high-effort generation loops. Crucially, the model retains its ability to ground its outputs in both standard Google Search and Google Image Search, allowing it to dynamically pull and reference visual and factual information before rendering a final pixel.

Alongside these creative enhancements, Google is implementing a major downward adjustment in developer pricing via the Google developer API, the core infrastructure that allows external companies and third-party applications to integrate the technology into their own software products. Under the new pricing structure, standard 1K resolution images generated through the API will cost $0.0336. This represents a precise 50% price cut compared to the previous Nano Banana 2 model, which charged $0.067 for the same output.

Higher-resolution generation sees a proportional reduction as well. A 4K image now runs $0.0756 through the API, down from $0.151 in the previous generation. Additionally, Google is introducing further incentives for high-volume users, offering an extra 50% discount for batch jobs—scenarios where large volumes of images are processed collectively in bulk queues rather than on-demand in real-time.

Calculated at the standard 1K pricing tier, producing one thousand images via the API now totals approximately $33.60, a steep drop from the $67 cost required under Nano Banana 2. By combining higher ELO preference scores, advanced multi-reference tracking, up to 4K native output, and slashed developer costs, Google’s latest release positions Nano Banana 2.1 as a powerful, highly efficient engine for the next wave of generative media applications.