Summary: Google Images marked its 25th year with a browseable homepage for US desktop users and image generation inside AI Overviews using the Nano Banana model. The homepage requires sign-in, is English-language, and rolls out over coming weeks; image generation also rolls out over the coming weeks in English and is limited to regions where AI Mode supports it. The article frames the shift as Search moving from finding images to making them.
The news: browsing and generating arrive together
Google Images reaches the quarter-century mark this week, and Google is marking it with a pair of updates that change how visual content behaves inside Search. First, Google Images gets a browseable homepage: a dynamic, immersive wall of pictures pulled from across the web, refreshed in real time and shaped around your interests. Ideas you save to collections while browsing show up as tabs above the main gallery, so returning to pick up where you left off takes no effort. Desktop users in the U.S. get it over the coming weeks, in English, and signing in to your Google Account is required.
The second update carries more weight for product builders: AI Overviews can now create images themselves. With Google’s newest Nano Banana model underneath, a plain text prompt now yields a high-quality custom visual built from scratch. This capability also lands over the coming weeks, in English, across every region where AI Mode already supports image creation.
Put simply, Search has stopped being only a place to find images and has become a place to make them. When the picture in your head doesn’t exist anywhere on the web, AI can draw it for you right on the results page.
The road here: a quarter century of milestones
Brad Kellett, Google’s Senior Engineering Director for Search, takes the story back to 2000, when Jennifer Lopez’s green Versace dress broke the internet. The search page of that era—a bare list of blue text links—couldn’t deliver what people actually wanted. They didn’t want another article about the dress; they wanted to look at it. Google Images filled that gap in July 2001, and visual content became searchable for the first time.
The milestones since then tell their own story:
- 2009: Similar Images – Made the picture itself the search clue. Look up “bow” and you might get hair bows next to bows and arrows; pressing “find similar images” sharpened the results with no fresh query needed.
- 2011: Search by Image – Let an uploaded photo or a pasted URL act as the query, so you could hunt down where an image came from and surface visually similar material.
- 2018: Google Lens in Search – Placed a search box behind your phone’s camera, identifying objects, translating text, and calling up product links on the spot.
- 2022: Multisearch in Lens – Crossed into multimodal territory by combining text and images in one query—photograph a landmark and ask “what inspired this design?”
- 2024: Circle to Search – Handed Android users four gestures—circle, highlight, scribble, tap—for searching whatever sits on their screen without hopping over to another app. Over 580 million Android devices have it today.
- 2025: Lens + AI Mode – Joined Lens’s multimodal muscle with AI Mode’s reasoning and applied a “visual image fan-out” technique that fans a lone image query out into dozens of sub-queries so the full context comes into view.
- 2025: Search Live – Brought your phone’s live camera feed into an interactive voice conversation in AI Mode, taking in movement and everything around you.
- 2025: Visual Results in AI Mode – Created a fresh route to exploring, imagining, and shopping: describe what you want in conversation (“barrel jeans that aren’t too baggy”, for instance) and receive a grid of visual inspiration next to shoppable products.
- 2026: Circle to Search Multi-Object Recognition – Taught Circle to Search to work through several objects in one image at once, again leaning on visual image fan-out.
- 2026: Intelligent Search Box – Delivered the Search box’s biggest overhaul in over 25 years: upload images—several at a time if you like—ask detailed questions about them, and get answers powered by AI Mode.
Lay the timeline out and one trajectory is unmistakable: text gives way to images, static gives way to dynamic, and retrieval gives way to generation.
What product teams should take from this
Teams building AI products can act on several concrete lessons:
The ways you can ask keep multiplying. Search input has climbed a ladder: text first, then images, the camera, circling gestures, live video, and today prompts that generate. A product built on a single input modality should think about welcoming more natural, multimodal exchanges. The direction of travel is meeting users wherever they turn up, be it a snapshot, a gesture on screen, or a live feed.
Finding things and making things are merging. Wiring image generation straight into AI Overviews recasts Search as a content production tool, no longer just an information retrieval system. Products that live on search traffic should take note: when people can conjure the exact image they want rather than clicking through to somebody else’s website, that traffic stops behaving as it did. Ask what your product gives people once pre-existing content alone is no longer the draw.
Real-time personalization is turning into the baseline. The browseable homepage leans on constant refreshes and tailoring to individual interests, making visual exploration feel like a living gallery rather than a fixed list of results. If visual content is your product’s territory, discovery should feel this alive and this personal.
“Visual image fan-out” is a pattern worth borrowing. Splitting a single picture into dozens of sub-queries so a system can understand its context translates well to any AI product that has to make sense of intricate visual scenes. The point isn’t naming objects; it’s reading relationships and intent.
Caveats worth weighing
New features come with fine print:
- Where and in which language: Everything rolls out in English, and image generation reaches only regions where AI Mode already supports image creation. The browseable homepage stops at the U.S. for now, so non-English markets and enterprise users will probably wait.
- Sign-in required: The homepage asks for a Google Account, which shuts out logged-out sessions and privacy-minded users.
- The generative layer is still experimental: Google labels the post’s audio “generated by Google AI” and carries the notice “Generative AI is experimental.” The broader AI features share that caveat—expect the occasional inaccuracy or odd output.
- Quality rides on Nano Banana: Image generation is only as good as the model beneath it, which brings its own limits in style, accuracy, and safety filters.
For builders, the read is simple: don’t assume day-one global coverage. Plan a gradual rollout and decide what your product does in the regions and languages these features haven’t reached.
One lesson to act on: build for the “see” moment
Across 25 years of visual search, innovation keeps landing wherever users need to see rather than read. A dress people wanted to look at produced Google Images. Similar Images cut through the trouble of describing pictures in words. Lens made the physical world searchable. And image generation in AI Overviews now covers the moment when the right picture doesn’t exist anywhere yet.
If visual AI is on your roadmap, open with this question: “When does my user need to see something that they can’t easily describe in words, or that doesn’t exist yet?” That’s where search and generation add real value.
Concretely: an e-commerce app could render the product in your chosen color or setting before you buy. A design tool could turn a described concept into a first draft to iterate on. A travel app could show a destination across seasons and times of day. Find the “see” moment in your users’ journey and strip the friction out of it.
Google’s own arc maps the route—first make visual content findable, then interactive, then generative. Your product’s next step along that path may be closer than it looks.