Google upgrades AI to create ‘realistic’ photos

According to Google, Nano Banana 2 takes advantage of the latest Gemini 3.1 Flash Image model. This image creation AI still retains the ability to produce detailed and high-quality images similar to the previous Banana Pro with faster time, with resolutions ranging from 512 pixels to 4K at different frame rates.

Besides, Nano Banana 2 also follows command prompts better, reduces content errors, can maintain consistency for up to 5 characters and detail of 14 objects in a single workflow, helping users “tell stories through photos” better. Google says users can request complex nuances of detail to create photos or multimedia products with vivid lighting, rich textures and sharper details.

Nano Banana 2 will become the default imaging model across all apps in the Gemini suite, both free and paid, as well as in the Flow video editing tool. On Google Search, Nano Banana 2 also defaults to photo search results in Google Lens and AI Mode, deployed in 141 countries globally for both PC and mobile devices. For developers, Nano Banana 2 will be available through Gemini API, Gemini CLI and Vertex API.

Google said that all images created will have a SynthID embedded – used to verify whether image, video or audio files were created or edited by an AI tool. According to Google, since the launch of SynthID in November 2025, users have used it more than 20 million times.

In addition, the resulting photos also meet C2PA content certification – an open technical standard that allows publishers, companies and third parties to embed metadata into media to verify origin, backed by a coalition of companies including Adobe, Microsoft, Google, OpenAI and Meta.

Google first released Nano Banana in August 2025, upgraded Nano Banana in November of the same year, allowing users to create fast and high-quality AI images. The product currently competes with a series of equally powerful tools from xAI Grok, Midjourney, OpenAI Dall-E, Stable Diffusion or Chinese platforms such as Alibaba’s Tongyi Wanxiang, Baidu Ernie, Tencent Hunyuan Image or ByteDance Jimeng.