Celebrating 25 years of visual search innovation

In 2000, Jennifer Lopez's green Versace dress set the internet ablaze; users were no longer satisfied with text links—they wanted to 'see' that photo. This demand directly gave birth to Google Images, which launched in July 2001, shifting search from blue text links to visual presentation. Twenty-five years later, Google Images not only lets users see the world but also enables them to 'create' it.

In July 2026, the official Google Blog announced two new features: a dynamic, browsable homepage that pushes real-time images based on user interests and supports favorite tag management (first launched on the US English desktop version); and image generation integrated into AI Overviews, where users input text prompts and the Nano Banana model generates new images. This marks the shift from searching existing information to generating new content.

The evolution of visual search did not happen overnight. The 'Similar Images' feature in 2009 allowed users to get similar results without text; 'Search by Image' in 2011 made the image itself a query; Google Lens in 2018 turned the camera into a search entry point; Multisearch in 2022 enabled combined text+image queries.

Circle to Search in 2024 embedded search into any scenario through gesture interaction, now covering 580 million Android devices. In 2025, Lens combined with AI Mode used 'visual image fan-out' technology to break a single image into dozens of sub-queries, and introduced Search Live (video conversational search) and AI Mode visual results. In 2026, Circle to Search's multi-object recognition allows one circle to identify all elements in the frame. Each step lowers the search barrier and raises the perceptual dimension.

The core of the browsable homepage is a 'dynamic immersive gallery' that adjusts content in real time based on the user's logged-in behavior and collection preferences. Saved images appear as tags above the main image for easy access—essentially a personalized visual news feed. However, its initial scope is limited: only US desktop English users, and requires a logged-in Google account.

More groundbreaking is the image generation in AI Overviews. Google describes this capability as 'turning text prompts into high-quality custom visuals from scratch,' powered by the Nano Banana model. This continues the logic of 2025's AI Mode visual results (e.g., descriptive search for 'non-puffy bootcut jeans'), but expands from 'finding' to 'generating.' However, this feature currently works only in English and in regions where AI Mode is supported.

From retrieval to generation, Google Images's 25 years reflect a shift in the core logic of search products: initially helping users find existing information, now helping users create content that doesn't yet exist. This shift is also evident in tools and interactions: search is no longer just a search box but a combination of camera, gestures, and multi-turn conversations.

But the new features also imply limitations: image generation depends on the model's capabilities; the Nano Banana is Google's latest model but may bring issues of copyright, quality, or bias. Additionally, the personalization algorithm of the browsable homepage could reinforce filter bubbles. The official Google Blog did not mention these risks, and all descriptions maintained a positive tone.