AI Visual Search

You rarely remember what a file was called. You remember what was in it. AI Visual Search reads the actual content of your photos and videos, so you can describe what you're after in plain language and get it back.

When you type a query into the main search bar, Tagbox runs two searches simultaneously:

This means you don't need to choose between the two. A single search covers everything, so you're less likely to miss a file.

Our search understands meaning, not just words, so three quite different kinds of query all work:

Search leans towards showing you more rather than less. A file that only partly matches still comes back, because a near miss you can see beats a perfect match you never find.

Filters and tags work the other way round. Both are exact: ask for landscape and you get landscape, nothing else. Reach for search when you're describing something, and for filters and tags when you already know precisely what you want.

AI Visual Search is now available for video files, in addition to images.

When Tagbox finds a visual match inside a video, clicking on that video doesn't just open it - it takes you directly to the first position in the video where the match was found. You can also see where matches appear across the full video in two places:

This makes it easy to jump straight to the relevant moment without scrubbing through the entire clip.

For a full list of supported video formats, see Adding content to Tagbox.

Search results use color coding to indicate how strong each visual match is:

Both types of matches are shown to give you the widest possible set of results, while still making it easy to prioritize the most relevant ones. Start with the bright yellow results and work your way through the faded ones if needed.

Sometimes the fastest description is the picture itself. Image search sits just to the right of the search bar: upload a picture and you'll get everything in your library that looks like it, closest match first. Handy when a client sends a reference and you want to know whether you already have something like it.

You can also open any file and use see similar to pull up its closest relatives.

Results come back sorted by how well they match, best first. There's no sort control to change that, and there doesn't need to be - if what you want isn't near the top, the fix is a better description rather than a different sort order.

Search follows you. Run it from inside a collection and it searches that collection rather than your whole library, which is usually what you want once you've already narrowed things down.

Every search updates the address bar, so the way to keep one is to bookmark it. Send that link to a colleague and they land on the same results, with your filters and tags still applied.

See the full tutorial below: