Google has confirmed that the mysterious AI model nano-banana that has been wowing LMArena users is its own. It is the basis for an update to its Gemini chatbot for image generation and processing, giving it precise control over photo editing.
Google has confirmed that the mysterious AI model nano-banana that has been wowing LMArena users is its own. It is the basis for an update to its Gemini chatbot for image generation and processing, giving it precise control over photo editing.
The update, called Gemini 2.5 Flash Image, will be available starting Tuesday to all users of the Gemini app, as well as to developers via the Gemini API, Google AI Studio, and the Vertex AI platform, TechCrunch notes.
The new Gemini AI model is designed to more accurately edit images based on natural language user queries, while preserving the integrity of faces, animals, and other details that AI typically fails to do. For example, if you ask ChatGPT or Grok to change the color of a person’s shirt in a photo, they can output a changed face or background in response.
«Reuse the same characters, changing their outfits, poses, lighting, or setting. Or reinvent yourself—decades later, in different locations, or at your childhood dream job,» Google describes the new Gemini capabilities.
Gemini’s new features also allow users to upload up to three photos for AI to combine into one. With prompts, users can specify which objects, elements, colors, or textures from the references should be used in the final image.
The update also improves image editing capabilities. New tools allow you to change backgrounds, restore faded images, and change outfits. Users can initiate these changes using simple language prompts.
«We are really pushing the quality of visualization forward, as well as the ability of the model to follow instructions. This update significantly improves the editing process, and the results of the models can be used for any purpose,» said Nicole Brychtova, head of visual model generation at Google DeepMind.
According to her, the developers created this model based on real user cases, for example, to help them visualize their home and garden projects.
While its new AI-powered image generator Gemini makes it easy for users to create and edit realistic images, the company has safeguards in place that limit what users can do. Last year, Google was embroiled in a scandal when users complained that Gemini generated historically inaccurate images; the company was even forced to apologize.
«We use extensive filtering and data labeling to minimize harmful content in datasets and reduce the likelihood of harmful outcomes. We also aggregate and assess content safety, including child safety, and its representativeness,» Google noted in the description of the latest update.
The company warned that Gemini now has all the latest privacy and security features, including SynthID, which embeds an invisible digital watermark directly into an image, allowing it to be identified as being created by artificial intelligence.
«We want to give users creative control so they can get what they want from the models. But that doesn’t mean everything is allowed», — said Brichtova.
Additionally, in its terms of service for generative AI, Google prohibits users from creating «intimate images without consent».
