Turn a broad search into a shot you can use

A search for “work” can match many kinds of footage. This exercise turns that broad idea into a concrete visual request, then uses playback to decide whether a result is useful. The goal is one reviewed source shot, with its file and detected time range checked.

Use a small collection you know well. For example, choose office footage that includes people, desks, and meetings. The queries below are examples to adapt to your own library.

1. Check that you are searching the right material

Select the project collection in the search filters. Confirm that its shots have completed AI understanding. A video appearing in the library is not enough: local import and AI processing are separate steps. See Import and AI understanding if the material is still being processed.

Check the search mode or any status message. Semantic search and exact filename or tag search answer different questions. If ShotAI reports a fallback, resolve that condition before judging the semantic results. The search guide explains the two modes.

2. Replace the theme with a visible action

Start with “people working.” Open a few promising results and decide what is missing from your request. Perhaps you need collaboration rather than someone working alone. Change the query to “colleagues discussing ideas around a table.”

If you need a more specific image, add one visible detail, such as “colleagues discussing sticky notes on a glass wall.” Change one requirement at a time so you can understand which wording changes the candidates. Avoid combining unrelated scenes into a single request.

3. Widen deliberately when results are missing

If nothing useful appears, check the collection and processing status first. Then remove a detail you can compromise on. For example, try “colleagues discussing sticky notes” before dropping the action entirely.

Use available filters when the requirement is about the collection or duration. Do not treat an empty result as proof that the footage does not exist. Equally, a returned candidate is not proof that every detail is present. Troubleshooting provides the next checks when changing the wording does not help.

4. Verify the result by watching it

Open the shot and review the full returned range. Confirm the people, action, setting, and any other essential condition. A thumbnail or summary can omit a change that happens later in the shot.

Treat the relevance score as a ranking aid, not a probability that the answer is correct. The returned time range belongs to the detected shot; it may include footage before or after your desired moment. Plan any final trim in your editing software.

5. Keep a useful reference

Add the chosen whole shot to the current selection. Check its source filename and start/end range in Selected Shots. Keep those details in your own project notes if you need to return later; do not depend on the current selection as a saved, named project list.

You are finished when you can explain why the shot matches the request and reopen the source to verify it. Continue with Review and export when you are ready to use it.