જેમિની 2.5 ફ્લેશ ઇમેજ

Gemini 2.5 Flash Image, known as Nano Banana, is a state-of-the-art image generation model with real contextual understanding. It handles generation, edits and multi-turn conversations about an image, and supports aspect-ratio control on every request.

This is the model that started the line, and the one everything since has been measured against. Google built it as an answer to specific feedback on its previous native image generation: developers liked the latency, the cost and the simplicity, and wanted higher quality and real creative control. What arrived added four things at once - blending several images into one, holding a character's appearance across prompts and edits, making targeted changes in plain language, and bringing the language model's knowledge of the world to bear on both generation and editing.

Character consistency was the headline, because it was the standing problem. Keeping a subject recognisable across environments and angles is what turns single images into a set: the same person in different scenes, one product photographed from several sides, brand assets that match each other. Google extended the same idea to templates, and pointed at property listing cards, uniform employee badges and whole product catalogues generated from a single design.

The editing model is conversational rather than tool-based. Blur a background, remove a stain from a shirt, take a person out of a photo, change a subject's pose, colourise a black-and-white image - each is a sentence, not a selection. Multi-image fusion is the same interaction applied to more than one input: drop an object into a scene, or restyle a room to a colour scheme or texture, in one prompt.

Pricing was quoted per output token, at a fixed number of tokens per image, which worked out to roughly four US cents a picture and set the reference point the rest of the family has been priced against. Every image carries an invisible SynthID watermark identifying it as AI-generated or AI-edited.

Google named the limits at launch and has since named a bigger one. At release the lab said it was actively working on long-form text rendering, more reliable character consistency and faithful fine detail - which is exactly the list the later Pro and Flash image models were built to close. Google now describes this model as the legacy member of the family and recommends the current Lite model as a direct replacement, on the grounds that it is better, faster and cheaper. Keep this one for a pipeline already calibrated to its output.