How VidFinder Search Works: Two Engines, One Search Bar

VidFinder runs two search engines behind one search bar. Standard search is always on and matches the words attached to your footage – file names, folder names, and tags. Smart Search is an optional AI layer that matches what a clip looks like, even when nothing is named or tagged. You type in one place, and VidFinder decides how hard to look.

Standard search is the wide net. Smart Search is the part that reads the picture. Most of the time you never think about which one is running – you type, results appear. The rest of this page explains what each engine does, so you can steer the search bar on purpose instead of guessing.

When you type a word, standard search checks three places at the same time: the file name, the full folder path the clip sits in, and any tags you have added. If your word shows up in any of the three, the clip appears. It matches partial words too, so bea surfaces everything named or tagged beach before you finish typing.

That three-way match is deliberate. It means you never miss a clip because you tagged it one way and named it another. A file called C010C023_260610FK.mp4 sitting in /Shoots/Camera A/ with a Hexagon tag can be found by its name, its folder, or its tag – whichever you happen to remember. Case and punctuation do not matter: Cam A, cam a, and cam-a all do the same thing, because the search keeps only the letters and numbers and ignores everything else.

Every word you add narrows the results, because a clip has to match all of them. Start with one word to cast a wide net, then add a second to tighten it. Cam A returns everything related to Cam A. Cam A Hexagon returns only the clips that match both. Cam A December narrows further, down to a single shoot.

This is the core move: start broad, then stack. Say you are cutting a brand video and you need the wide shots of Camera A from the Hexagon project. Type cam a hexagon wide and standard search requires all four terms, handing you exactly that set instead of the whole camera roll. Results follow your Sort order – newest first by default – rather than a relevance score, so stacking a term is how you cut a long list down, not re-rank it.

Standard search reads folder names and file names, not just tags. So a clip can appear for a term it was never tagged with. Every clip inside a folder called "Camera A" shows up when you search Cam A, even the ones you never touched. That is the wide net doing its job, not a bug.

If you want only the clips carrying one specific tag, the fastest fix today is to stack a second term that narrows the set. A dedicated tag-only filter is on the roadmap for when you want strict tag matching. Until then, remember the rule behind the surprise: any untagged clip in your results matched through its file name or its folder path, never through a guess.

Smart Search uses an on-device AI model to read what is visible in each clip, so you can describe a shot in plain words and find it with zero tags and a camera-gibberish filename. Type sunset over water, person in a red jacket, or close-up of hands typing, and it ranks clips by how closely each one matches your description.

It matches against the whole clip, not just the thumbnail. VidFinder samples 8 frames spanning the full length of each video – evenly spaced from 5% to 84% of the runtime – so a moment two thirds of the way in still counts. A client asks for "that shot of the coffee being poured." You never tagged it, and it is named A001_C012.mov. Type pouring coffee and Smart Search finds it by sight.

Smart Search matches what a frame looks like, not what is said or written on screen. It reads the picture, so it will not find spoken words and it will not read on-screen text – it is visual, not a transcript search. Because it samples 8 frames per clip, a detail on screen for a split second between those samples can be missed.

There is also a one-time setup cost worth stating plainly. Smart Search needs a model download of about 149 MB the first time. You do not have to trigger it: once you are signed in, the model arms itself in the background, and existing libraries backfill on their own. Standard search works the whole time, with or without the model.

Every part of Smart Search runs on your machine. The AI analysis, the frame sampling, and the search all happen locally. Your videos, the extracted frames, and the search index are never uploaded. The only thing VidFinder ever downloads is the AI model itself, one time. Your footage stays where it already lives.

For anyone working with unreleased or confidential footage, that is the point. There is no cloud step to trust, no upload to a third-party server, and no copy of your library sitting somewhere else. The model comes down once; nothing about your footage ever goes up.

A few habits make both engines faster to drive:

• Start broad, then stack. One word to cast a wide net, a second to narrow. Every added term tightens the set.
• Partial words work. pro finds product, promo, and proposal – type only as much as you need.
• Case and punctuation do not matter. Cam A, cam a, and cam-a behave identically.
• Name your folders well. Folder names are searchable, so a tidy folder structure makes everything easier to find.
• Describe the shot, not the file. When you cannot remember what you called it, switch to Smart Search and describe what is on screen.