The Mechanics of Google Circle to Search: Multi-Object Recognition

AI

📝 Key Points of This Article

• Google explains the technology behind the new “multi-object image recognition” feature for Circle to Search.
• The system has evolved from single-object recognition per image to recognizing multiple objects within a single image.
• This is made possible by a core technology called the “Fan-out” process.

In late February 2026, Google announced a new feature for the Android gesture-based search tool “Circle to Search” called “multi-object image recognition,” and began rolling it out to the Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, Pixel 10 Pro Fold, and the Samsung Galaxy S26 series.

Circle to Search has evolved from single-object recognition per image to identifying multiple objects within a single image. This advancement is naturally driven by the evolution of the underlying AI model, “Gemini.” By introducing a more advanced Gemini AI model, the technology can now analyze images alongside user queries, enabling sophisticated multi-modal reasoning.

Related Article

かこって検索「複数オブジェクト画像認識」の裏側

Google、「かこって検索（Circle to Search）」の新機能として展開を開始した「複数…

jetstream.blog

Until now, Google image search tools like Circle to Search and Google Lens were primarily used to search for one item at a time—for example, selecting a specific pair of sneakers from an outfit seen on social media, or searching for a lamp in a photo of a stylish living room. However, the latest AI model can now interpret an entire image as a complex scene, break it down into individual components, and identify multiple items such as a hat, jacket, and shoes simultaneously. It then compiles the search results for these items into a single, easy-to-understand response in one action.

The core technology supporting this simultaneous multi-object recognition is a process called “Fan-out.” The moment a user initiates an image search, Gemini infers the user’s intent while identifying multiple objects in the image. It then executes multiple searches in parallel, branching out like a fan to retrieve results from billions of web entries.

The AI then organizes this vast amount of information and returns it to the user in a matter of seconds. This is what is happening behind the scenes of the new multi-object image recognition feature in Circle to Search.

Furthermore, this “Fan-out” process is not limited to Circle to Search and Google Lens; it is also active in prompts used within the Gemini app and the AI search service “AI Mode.”

Source：Google