Google introduces Nano Banana 2.1 with cheaper image generation

Google has released Nano Banana 2.1, cutting image-generation costs by approximately half while improving editing accuracy, visual consistency and text rendering, as developers face a deadline to migrate from its predecessor.

The new artificial intelligence model, introduced on 6 October, targets developers, businesses and creative professionals using automated image production. Google says the upgrade delivers stronger visual quality, better adherence to instructions and greater consistency when users make successive changes to generated images.

Nano Banana 2.1 is available through Google’s Gemini API and AI Studio, alongside its broader image-generation ecosystem. The company has positioned the model as its principal high-efficiency option for producing and editing images, combining improved capabilities with lower processing costs.

Under Google’s published API pricing, generating a standard 1K image costs approximately $0.0336, compared with $0.067 for Nano Banana 2. A 2K image costs $0.0504 against approximately $0.101 previously, while 4K generation is priced at approximately $0.0756, down from $0.151.

These figures cover image-output charges rather than the complete cost of individual requests. Input processing, text generation, internal reasoning and certain search functions can attract additional charges, depending on how developers configure their applications.

Google also offers batch processing at lower rates, bringing the image-output cost to approximately $0.0168 for 1K, $0.0252 for 2K and $0.0378 for 4K. Batch processing is intended for workloads where immediate responses are unnecessary.

The pricing changes could benefit companies generating large volumes of marketing materials, product illustrations, advertising designs and digital content. However, Google’s higher input-token charges mean the overall savings will vary according to the complexity of individual requests.

Developers using Nano Banana 2 through the Gemini API also face a migration deadline. Google’s published retirement timetable identifies 29 October as the shutdown date for the previous model’s API endpoint, leaving approximately three weeks between the new release and the scheduled transition.

Applications relying on the older model identifier, gemini-3.1-flash-image, will therefore need adjustments to continue operating through the replacement service. Google identifies gemini-nano-banana-2.1 as the new model identifier.

The upgrade introduces improvements to conversational image editing, allowing users to refine an image through successive instructions while maintaining greater consistency in subjects, objects and composition.

Such capabilities address a persistent difficulty with generative imaging systems, which can unintentionally alter facial features, clothing, backgrounds or other details when users request seemingly minor modifications.

Google says Nano Banana 2.1 improves character consistency across multiple editing steps and strengthens adherence to instructions involving several visual elements.

The model supports up to 14 reference images within a single generation workflow. Google’s technical documentation specifies support for maintaining consistency across as many as four characters and preserving the visual identity of up to 10 objects.

Text rendering has also been improved, particularly for designs involving typography, infographics and structured visual layouts. The company says the system produces more accurate arrangements of written content within generated images.

Another improvement addresses visual artefacts affecting unusually wide or tall images. Google says it has corrected tiling problems encountered with panoramic aspect ratios, including 4:1 and 8:1, at higher output resolutions.

The model supports 1K, 2K and 4K output, enabling developers to select image quality according to their intended applications and processing budgets.

Google has incorporated search grounding capabilities, allowing the system to draw on information from Google Web Search and Image Search when generating visual content requiring external references.

Developers can also configure the model’s reasoning effort through minimal, medium and high settings. Medium is the default, although different settings can influence processing requirements and associated charges.

The release strengthens Google’s range of image-generation products, which also includes Nano Banana 2 Lite for lower-cost workloads and Nano Banana Pro for more demanding visual tasks.

Google’s enterprise documentation confirms that Nano Banana 2.1 became generally available on 6 October, with support for batch inference and several enterprise security controls.

The model accepts text, images, video and PDF inputs, while producing text and image outputs. Its published technical specifications list an input-token limit of 131,072 and an output-token limit of 32,768.