Google Introduces Gemini Nano Banana, a New AI Image Model With Improved Character Consistency

Google introduced a new artificial intelligence (AI) image model on Tuesday, calling it the company's best image generation and editing model to date. Dubbed Gemini 2.5 Flash Image, the model improves on both the generation speed and element-based editing. Interestingly, the model went viral on the crowdsourced AI model ranking platform LMArena weeks before its official announcement. While in stealth mode, the model was named Nano Banana, and it has received praise from users for its high-quality image editing and consistent character. Notably, the Gemini 2.5 Flash Image is now available within the Gemini app.

Google's New Image Model Offers Improved Editing Capabilities

In a blog post, the Mountain View-based tech giant admitted that the Nano Banana AI model, which recently ranked first on LMArena, was in fact Google's Gemini 2.5 Flash Image. The company is not the first to test its model in stealth mode, as OpenAI was recently also spotted testing its GPT-4.1 model in a similar manner.

While the company claims state-of-the-art (SOTA) speed and quality in image generation, the new model's key ability lies in image editing. Google says the model can now maintain a higher character consistency while editing elements within the image.

One issue users have faced while editing non-synthetic images using Gemini is that the person in the frame sometimes gets distorted or altered to appear nothing like the input. Gadgets 360 staff members tried out the new model in the Gemini app and found that it is now able to change things like the colour of the t-shirt or add a hat to a person without affecting the person in the frame.

The Gemini 2.5 Flash Image also gets a new image blending feature. With this, users can now take two different images and ask the AI to blend them into a single image. The success rate is mixed, and it largely depends on both the input images and the prompt. However, it is a fun tool to try out.

Finally, the new image model also allows users to make multi-turn edits. This means that if users are not satisfied with the first iteration, they can continue requesting more changes in subsequent prompts, all while keeping the base character consistent.

While end consumers can access the model in the Gemini app, developers can access it via the Gemini application programming interface (API), Google AI Studio, and Vertex AI (for enterprises). It is priced at $30 (roughly Rs. 2,600) per million output tokens, with each image consuming 1290 tokens. This means each image generation will cost $0.039 (roughly Rs. 3.5).