Google has unveiled a powerful enhancement to its Gemini app by integrating a new AI image‑editing model, codenamed Nano Banana, officially branded as Gemini 2.5 Flash Image. The update empowers users with tools to consistently preserve likeness in multi‑step edits while opening up creative possibilities with unprecedented realism and control.
Users can now instruct Gemini via natural language to modify images—whether it’s changing hairstyles, adding props, or altering backgrounds—without compromising the identity or visual consistency of subjects such as people, pets, or objects. This “character consistency” addresses a long‑standing criticism of AI editors, where subtleties often shift between edits, producing results that feel “close but not quite the same.”
Beyond preserving likeness, the tool introduces multi-turn editing, enabling users to engage in iterative spot‑on adjustments—in effect, editing a specific part while the rest remains intact. Users might start with an empty room, then sequentially paint the walls, add furniture, and fine‑tune décor without compromising earlier changes. Moreover, the design mixing feature allows one to apply aesthetic elements from one image, like a texture or pattern, to an object in another—say, transferring butterfly‑wing motif onto a dress or using floral textures to style rain‑boots.
Interoperability marks another leap forward. Users are now able to merge multiple images—like combining a selfie with a picture of a pet—into inventive new compositions while maintaining fidelity to each subject’s appearance.
In terms of accessibility, Nano Banana is available globally to both free and premium users of the Gemini app across web and mobile. Developers can also leverage the model via Gemini API, Google AI Studio, and Vertex AI platforms. Images generated or edited through Gemini—which employ SynthID technology—carry visible watermarks as well as embedded digital identifiers to signal AI origin.
Google’s ambition is clear: to refine user control and visual fidelity in AI‑powered imagery. According to Nicole Brichtova, a product lead at Google DeepMind, the model raises the bar for seamless editing and instruction‑following, producing outputs that are “usable for whatever you want to use them for.”
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
