What AI search can and can't do

Search in Total Vault is built to understand what you actually mean, not just match filenames. You can describe what is in a photo in plain language, narrow by when it was taken, and point the search at a single project or at everything you've uploaded.

It is genuinely good at this today. But it is honest about its limits too, so this article lays out both sides: what works now, and what is still on the way.

Type a description the way you'd say it out loud. Search reads your words, figures out what you're looking for, and ranks results by how well each file matches.

• Find photos by what's in the frame. Try sunset on the beach, ring close-up, people dancing, or bouquet. Search looks at the picture itself, not just its name, so it can surface the right shot even when the file is called something like DSC_0481.
• Search in plain English. You don't need keywords or special syntax. A normal sentence like candid photos from the reception works fine.
• Narrow by date. Phrases like in 2024, last year, this year, or recently are understood and used to filter by when a photo was taken. You can also set a date range from the Modified filter on the results page.
• Scope to one project. Say in the Sundays project only and search will keep results inside that project, if a project by that name exists. You can also run a search from inside a project to look there first.
• Search across everything. By default, search looks across every file you've uploaded into your workspace at once, so you don't have to remember which project something landed in.
• Filter by file type and source. On the results page, narrow to photos, videos, RAW, documents, and more, or to how a file arrived (a direct upload versus a Guest upload link).

When you search, Total Vault does a few things at once and blends the results, so a great match can come from any angle:

• It looks at the visual content of your photos to find ones that look like what you described.
• It checks the AI tags that were added to each file when you uploaded it.
• It matches the words in filenames, captions, and other text.

Those signals are combined and ranked together, which is why a search can find the right photo even when no single field is a perfect match.

Being honest matters more than overpromising. Here's where search will fall short today:

• It can't recognize specific named people by their face. There is no face recognition yet, so searching for a person by name won't reliably gather every photo of them. You'll see a People filter on the results page marked as coming soon, with a note that it'll let you scope by who uploaded or shared a file.
• Date filtering works at the year level. A phrase like May 2026 is understood as the whole of 2026 rather than just that month, so expect year-range precision, not month-by-month.
• Results depend on what's been tagged. Visual and tag-based matching rely on a file having been processed after upload. A brand-new upload may take a moment before it shows up for a content search, and very unusual subjects may match less confidently.

• Describe the picture, not the filename. Two people laughing outdoors will beat guessing at a file number.
• Add a date or a project when you know it. Beach photos from last year or portraits in the Riley project narrow things fast.
• Start broad, then filter. Run a simple search, then use the Type, Modified, and Source filters on the results page to tighten it.