Google is testing a new image AI model within its Gemini portfolio under the name Nano Banana 2 Flash.
Google is said to be positioning Nano Banana 2 Flash explicitly as an alternative to heavier image AI models, with a focus on fast output and lower costs. That’s according to BleepingComputer.
That Google is actually testing the model is evident from mentions that have surfaced on X and been shared by MarsForTech. This is a source that has previously released accurate information about unannounced Gemini models. This indicates that Nano Banana 2 Flash is already in an advanced testing phase, rather than being a pure concept or in early development.
Nano Banana 2 Flash will be part of Gemini’s Flash series, which is designed for applications where latency and scalability are more important than maximum complexity. This is also expected to apply to image generation. The model is said to be faster than Nano Banana Pro and also cheaper to deploy, but it does compromise on power and advanced reasoning capabilities.
Nano Banana 2 Flash is based on Gemini 3 Flash. It therefore stands to reason that the model focuses strongly on the efficient use of computing power. Sources suggest that the image quality is close to that of Nano Banana Pro, although the model is less suitable for complex creative tasks that require accuracy and deep contextual understanding.
Demanding creative applications
Currently, Nano Banana Pro, also known as Gemini 3 Pro Image, is Google’s most advanced image AI model. This model was developed for demanding creative applications such as creating diagrams, infographics, storyboards, and other visual concepts where precision and consistency are crucial. Thanks to more extensive reasoning capabilities and broader knowledge, it can also better handle complex prompts and reference material.
Nano Banana 2 Flash, on the other hand, seems primarily intended for fast image generation and editing at scale. This makes the model potentially interesting for developers and organizations that want to generate large amounts of visual content without the cost and latency of a top-of-the-line model.
Although Google has not yet made an official announcement, the fact that the model is already being tested suggests that a wider rollout is imminent.