Exclusive: New leaked image samples by Nano Banana 2

Following the brief appearance of Nano Banana 2 access on Media.ai, the model’s preview was quickly removed after image samples circulated widely across social media. These visuals highlighted not just the technical progress, but also the rising anticipation among creators and industry observers. Alongside the Media.ai leak, new image samples attributed directly to Gemini further confirm that Nano Banana 2 delivers comparable performance, particularly in remastering low-quality images and addressing complex instructions for coloring and image editing.

Images from MarsEverythingTech 👀

Images from Leo 👀

A notable advancement is the model’s improved ability to interpret natural prompts and accurately follow instructions, especially when dealing with text. Examples show the model drawing complex text on a whiteboard with precise font consistency, an area where prior models struggled.

This leap in visual fidelity and instruction-following points to an increased level of world knowledge embedded in the model. The immediate relevance of these improvements extends to fields such as design, media production, and automated content workflows, potentially allowing users to generate banners and social media visuals programmatically via API. The efficiency gains for teams relying on rapid, high-quality visual content could be substantial if the model maintains its performance in production.

While there is still no official confirmation of the release date, the pace of leaks and early access incidents suggests the launch is imminent. This aligns with Google’s ongoing strategy of iteratively upgrading Gemini-powered tools and expanding their ecosystem for both creative and enterprise applications.