Google’s official description reads:
Google Image Swirl organizes image search results into groups and sub-groups, based on their visual and semantic similarity and presents them in an intuitive exploratory interface.
How it works
Search results are represented as stacks of images whereas the first image is fully shown thus giving a clue about the type of images grouped together. The stacks represent the different visual perspectives for a term (see the Apple query as an example). Once the user selects a stack, the images spread out in a relations diagram grouped by visual similarity. The original results are still visible on the side, making it easy to navigate back to the original query. This can be done repeatedly to drill down into the search result. The thickness of the edges connecting the images and groups represent the size of the connected result—the more images are in a subset the thicker the edge is displayed.
Google Image Swirl is still an early protoype and the available queries are limited, which means you can’t already search for any term. I see a lot of potential in the combination between diversity and similarity of search results as it let’s you gain overview as well as focus on a specific type of result. I recommend having a look at it and would love to hear your opinions about it in the comments.