Nano Banana 2 brings better visual consistency, more precise textual effects and even improved speed of execution.
After its update from Gemini 3 to Gemini 3.1, DeepMind is evolving its flagship image model. The Google laboratory, based in London, announces the arrival this Thursday, February 26 of Nano Banana 2, which should gradually replace its little brother Nano Banana Pro. Gemini 3.1 Flash Image, its technical name, combines, according to DeepMind, the speed of a flash model and the precision of a Pro model. The model, state of the art at Google, brings notable advances in visual consistency, control, and precision of textual output.
Knowledge of Gemini models
The real strength of Nano Banana models comes from the richness of the text and understanding of the model. While many other models are based on a diffusion architecture, Nano Banana is based on a Gemini model (in this case Gemini 3.1 Flash) with specific fine-tuning for image creation. Using a generic model makes it possible to reason about text and gives the AI the ability to “understand” what it generates. The model thus relies on Gemini’s knowledge and real-time search to generate specific topics with more precision. Nano Banana 2 provides more precise texts in the generated images. It is now possible to directly translate text into an image.
More so, the model adopts better subject consistency during image generation. It is possible to generate up to 5 identical characters and 14 objects across several generations of successive images, without the AI starting to change their physical details.
The model also manages to better capture the nuances in your initial prompt and thus reflect them more faithfully in the output. Nano Banana 2 also brings more natural lights, richer textures and sharper details (sharp feeling). Nano Banana 2 results can be generated in resolutions ranging from 512px to 4000px, like the Pro version.
Nano Banana 2 deployed simultaneously at Google
Nano Banana 2 is deployed on February 26 in the Gemini application, with Fast, Pro and Thinking modes. AI Pro and Ultra users, however, retain access to Nano Banana Pro. The model will also be deployed in Search (outside France), in AI Studio (including the API) as well as in GCP in Vertex AI and Gemini CLI. The Flow video generation template will also use the template to generate its visuals. At the time of writing, Google has not communicated an official price for its use via the API.
