Google turned Google Images into a generation engine on its 25th birthday. According to a July 14 blog.google post from senior engineering director Brad Kellett, the company is shipping a redesigned, personalized image gallery and adding AI image generation directly inside AI Overviews in Search. The announcement doubles as an anniversary retrospective and a signal of where visual search is headed next.
The first change replaces the plain results grid with what Google describes as a dynamic gallery, refreshed continuously and tailored to a signed-in user’s interests. Images saved to collections now surface as tabs above the main feed, so a user researching a kitchen remodel can return to that thread later instead of rerunning the same query. Google says the feature starts appearing on desktop within weeks, limited at launch to English-language users in the United States who are signed into a Google Account.
The second change is the bigger one. Google is building text-to-image generation into AI Overviews, using what it calls its newest Nano Banana model. A user who cannot find an existing photo matching a specific idea (the company’s own example was a bathroom repainted in dark green and butter yellow) can now generate one without leaving the results page. Google plans to bring this to more English-speaking markets over the following weeks, starting with the regions that already support image creation in AI Mode.
The framing matters more than the feature list. Standalone generators, Midjourney, OpenAI’s image tools inside ChatGPT, and Adobe Firefly among them, have spent two years training people to open a separate app whenever they want a custom image. Google is betting that most of that demand starts as a search query rather than a creative session, and that catching it inside AI Overviews removes the reason to leave Search at all. If the new gallery also turns Google Images into a destination people browse rather than a fallback when text search comes up empty, Google gains a retention lever that no dedicated image tool has: a personalized feed tied to a signed-in account.
For publishers and SEO practitioners, the announcement reads as a warning as much as a product update. If a meaningful share of image queries can now be satisfied by generation instead of retrieval, some volume that used to land on stock photo sites, retailer product pages, and image-heavy blogs may simply stop clicking through. Google’s post gives no figure for how much of current image search volume this could redirect, and it does not disclose whether generated images will be visually distinguished from indexed photographs inside the new gallery.
That last omission is the one worth watching. A gallery that mixes AI-generated images with real product photos, without a clear label distinguishing them, creates confusion for shoppers trying to judge what an item actually looks like, a problem retailers cannot fix from their end.
Teams that rely on image search traffic should audit which top-performing queries are generic enough to be satisfied by generation, color palettes, outfit ideas, layout inspiration, rather than a specific photo, and treat that segment of traffic as at risk over the next two quarters.
Google detailed both features in a blog.google post published July 14, 2026, credited to Senior Engineering Director Brad Kellett.
