Google rolls out faster Nano Banana 2 Lite image model

Google DeepMind has released Nano Banana 2 Lite, a new AI image model built for faster and cheaper generation across Google’s tools. Google says the model, formally named Gemini 3.1 Flash Lite Image, is aimed at rapid idea testing and prototyping where speed matters as much as output quality.

According to Google, the model is available in Google AI Studio, the Gemini API and the Gemini app through the Flash-Lite option. Developers can also use it through Google’s API, where the company says its lower compute demands reduce image-generation costs.

Built for speed and lower cost

Google says Nano Banana 2 Lite can create an image from a text prompt in about 4 seconds when using its default low-thinking mode. Ars Technica reported that the standard Nano Banana model takes about 20 seconds to generate the same images.

Google says the model averages $0.034 per 1,000 images through the API. The company lists API pricing at $0.25 for 1 million input tokens and $1.50 for 1 million output tokens, which Ars Technica reported is half the rate for Nano Banana 2.

Google’s higher-end Nano Banana Pro remains more expensive. Ars Technica reported that Nano Banana Pro input tokens cost $2 per 1 million, while output tokens cost $12 per 1 million, eight times the output price of Nano Banana 2 Lite.

Quality trade-offs remain

Google has presented examples comparing Nano Banana 2 Lite with its non-Lite image models, and the company says user ratings from Arena.ai’s Elo scores put the Lite model close to the standard versions. Ars Technica noted that broad preference ratings may not capture smaller defects that become clear after closer inspection.

Google says Nano Banana 2 Lite can struggle with text, especially when the lettering is small. The company also says infographics are more likely to contain incorrect data, and characters or people may appear less consistent across repeated generations.

Those limits mean Google’s Flash or Pro versions may still be better suited for single-image projects that require stronger text handling or more accurate photorealistic results, according to Ars Technica. For quick design variations or early-stage visual ideas, Google is positioning the Lite model as a faster and cheaper option.

Watermarks and wider Gemini access

Google says images generated with Nano Banana 2 Lite include SynthID watermarks, a system designed to label AI-made media even after edits. Ars Technica reported that broader access to low-cost image generation could increase the amount of AI-generated material circulating online.

Google also said it is expanding access to Gemini Omni Flash, a model it announced at I/O in May. According to Google, Omni remains limited to video generation and is now available in the Gemini API and Google AI Studio after earlier launches in the Gemini app and Google Flow.

This story draws on original reporting from Ars Technica.