Google’s ‘Nano Banana 2’ image model leaks spark frenzy online ahead of launch

Gemini nano banana

Google’s upcoming AI image-generation model, Nano Banana 2, is already creating a stir online, even before its official release. Images allegedly generated by the model have flooded social media platforms, showcasing a dramatic leap in realism, accuracy, and creative flexibility compared to its predecessor.

Originally introduced as part of Google’s Gemini 2.5 Flash rollout earlier this year, the first Nano Banana model gained traction for its ability to perform complex image edits and generate visuals from natural-language prompts. Now, according to leaks and social media chatter, Nano Banana 2 pushes those boundaries even further.

Tech tipster Testing Catalog News on X (formerly Twitter) claims that the new version excels at text rendering, infographics, world knowledge, charts, and following intricate user instructions, areas where earlier image models often stumbled. Reports also suggest that Google briefly listed Nano Banana 2 on internal or partner platforms before swiftly taking it down.

If leaks are accurate, Nano Banana 2 could launch as early as this week, bringing with it broader output options including 1K, 2K, and 4K resolutions, plus expanded aspect ratio support like 9:16 and 16:9. This would directly address one of the biggest complaints about the first Nano Banana, its inconsistent handling of image dimensions.

Social media users who claim to have accessed the new model are sharing striking examples of its capabilities. One notable breakthrough appears to be the model’s ability to generate analogue clocks showing accurate times, a detail most AI tools fail to replicate due to the 10:10 bias found in stock imagery. Others have shared full webpage mock-ups and UI screenshots rendered with near-perfect layout accuracy, suggesting a significant upgrade in spatial and visual understanding.

Some users even posted surreal compositions, such as images depicting US President Donald Trump, Ukraine’s Volodymyr Zelenskyy, and Russia’s Vladimir Putin on a vintage monitor display, a sign of how creatively flexible the new model could be.

There’s also speculation that Nano Banana 2 could be powered by Google’s next-generation Gemini 3.0 model, which is expected to unify multimodal reasoning across text, vision, and generation.