Google Search now generates AI images when it can't find what you're looking for on the web

Google is expanding the capabilities of AI Overviews in Search by adding on-demand image generation. When a user's query returns no suitable image from the web, Search will now synthesize one using a model called Nano Banana 2 Lite, drawing directly on the text of the search query to produce a relevant visual. The rollout is expected to begin within the coming weeks.

The move reflects a broader pattern in how AI Overviews have evolved since their introduction - shifting from a tool that summarizes existing web content to one that can generate original content when gaps arise. Adding image generation to that workflow means Search is no longer purely a retrieval system for this type of query; it becomes a generative layer that fills in where the indexed web falls short.

Nano Banana 2 Lite appears to be a lighter, more inference-efficient variant within Google's image generation model family, likely chosen for the latency and cost demands of integrating generation into a live search product at scale. Details on the model's architecture or training data have not been widely disclosed, but the "Lite" designation suggests it is optimized for speed rather than maximum output quality.

The practical implications are notable for both users and the broader web ecosystem. For users, it means visual answers become available even for niche, hypothetical, or highly specific queries that no photographer or illustrator has addressed. For the web, it raises questions about what role image hosting and stock photography sites play when a major search engine can synthesize on demand rather than link outward. How Google handles attribution, watermarking, or disclosure that an image is AI-generated will be worth watching as the feature reaches users.