AI Video Search: Find Shots and Dialogue in Your Footage

“Find the shot of the player in a red shirt” and “find the interview where someone talks about the launch” are different requests. One depends on the picture. The other depends on speech. Searching filenames won't necessarily answer either. AI video search helps you find footage by its contents. This guide is about searching your own production library, using Aspect's visual and spoken-content search. It explains how to inspect a result and get back to the original file so the search can lead to an edit. The illustrations below use a sample library.

Choose the information you want to search

What you remember	Search to start with	Example request
An action, subject, setting, or composition.	Visual content.	“A player kicking a ball while wearing a red shirt.”
A quote or the topic someone discussed.	Spoken content.	“Interview passages about preparing for the launch.”
A filename or part of one.	Filename.	“B001C009.”
A campaign, date, person, or maintained field.	The relevant metadata filter.	“Workshop footage from the autumn campaign.”

Semantic search means searching by meaning rather than only matching identical text. It helps when you remember the idea of a shot or passage but not its exact words. It also means you should inspect the result: related content is not always the content you wanted. Aspect's search documentation covers visual content, spoken words, configured people, filenames, tags, and metadata. Several of these can appear in the same request.

1. Start with a project you know

Choose a small, representative project. Include footage with visible action and a recording with speech. Make sure you have access to the files and allow the required analysis to finish after upload. A clip missing from an unfinished index tells you little about the quality of a query. Write down a few shots and quotes that you already know are present. Note the source files and approximate positions before you search. That gives you something concrete to check against the results. For the visual example below, the request describes a player, a ball, and a red shirt. Those are observable details. “The best moment in the campaign” gives the system much less to work with unless you explain what makes that moment useful.

2. Describe the visible shot

Open Search and describe what you want to see. Begin with the subject and action, then add details that would distinguish the shot from similar footage. For example:

A player kicking a ball while wearing a red shirt.

If the framing matters, add it: a wide shot, a close-up, or a view with room for a title. Avoid adding a detail you only vaguely remember as a firm requirement; it may exclude the footage you need.

Describe the content you want to see. This example searches the picture rather than relying on the clip's filename.

3. Inspect the filters and matching moments

Aspect translates the request into filters you can inspect and adjust. In the example, the Visual filter contains the scene description. Check that the system has interpreted the request the way you intended. If you included a date or project field, inspect that restriction too. A correct visual description combined with the wrong date can hide the right shot.

Check the Visual filter, then play the matching range.

Open a candidate and watch the matching passage with a little time before and after it. Check the detail that matters for the edit. A close-up of the right player may be a relevant search result while still being unsuitable for the wide shot you need. Treat the result as a candidate. Check whether its action, framing, duration, focus, and continuity work in your edit.

4. Search spoken content for an interview line

When you need dialogue, describe the quote or topic as spoken content:

Interview passages where someone explains how the team prepared for the launch.

If you remember the wording, use it. If you only remember the topic, try that first and listen to the candidate passages. Aspect's documentation supports both exact-phrase requests and descriptions of a spoken topic. Once you're inside a recording, open the Transcript tab. Its search field helps you locate a word within that file. In the example, “launch” appears twice. The match count and previous/next controls let you move between occurrences.

Search within an opened recording to locate the wording.

Selecting transcript text can take you to the corresponding spoken moment. Listen to the sentence in context before quoting it. Transcription can get names, numbers, or specialist terms wrong, and a sentence can change meaning when separated from its question.

Use the transcript to reach the passage, then listen to the recording.

5. Refine an ambiguous request

Suppose “the launch” returns product interviews, an event recording, and footage of a boat. Decide which part of the request needs clarification before trying a longer paragraph.

If you want dialogue, specify that someone is discussing the product launch.
If you want the event footage, describe the visible action or location.
If the results span several campaigns, add a maintained campaign field.
If nothing appears, remove a restrictive field and check the project, access, and indexing status.

Change one part of the request at a time so you can tell which change improved the results. For recurring needs, use custom metadata for stable facts such as campaign names and usage restrictions. A natural-language query should not have to guess information your team already knows.

6. Hand the source to your editor

Keep the source filename and useful time range with your selection. If the result is a quote, include enough context for the editor to judge its meaning. In the Aspect desktop app, right-click the result and choose Open in mount to reveal the original file on your desktop. The project needs to be available through your mounted filespace. Mounting a project provides the shared files through a drive that creative applications can access.

Open in mount connects the result to its original file. It does not by itself turn the matching range into a subclip in your editing application.

Open the file in your editing application and locate the selected passage. Confirm any timecode or elapsed-time differences before sharing a list of selects. A result shown as seconds from the beginning of a file may not use the same display as the source timecode in your timeline. If someone else needs to access the selection, confirm their permissions. Finding a file in your account doesn't establish that another editor or a guest can open it.

Check whether search works for your library

Use a small repeatable exercise instead of judging the system from one impressive result. Choose five to ten known shots, one known quote, and one request with no valid match. Include a silent clip and a shot whose filename doesn't describe its contents. For each request, record:

Whether the expected file and moment appear.
Whether the first few results are useful or merely related.
Whether the matching range includes enough of the action or sentence.
Whether you can open the original and use it in the edit.

Check missed shots separately from unfinished indexing or missing access. For the no-match request, inspect any suggested candidates rather than assuming that a returned result must satisfy the request. Keep the same footage and questions if you compare products.

Common mistakes

Using visual search for words someone said. A scene can show a launch event without containing the interview line you remember. Specify spoken content when the audio matters. Accepting the thumbnail as confirmation. Watch the passage. The relevant object might be visible while the action, timing, or framing is wrong. Expecting a named person to be recognized automatically. Aspect's people search uses the people your team configures with reference photos. Check that setup before depending on a person's name in a request. Treating a match as permission to reuse the footage. Review the maintained rights information and applicable releases. The image alone cannot establish the terms of use. If you're still choosing a platform, read our comparison of AI video search tools for production libraries. If you want to see the complete retrieval and editing handoff, bring a few representative clips to an Aspect demo.

FAQ

Visual search analyzes what appears on screen. It can search silent footage without relying on dialogue, provided that the footage has been indexed for visual search.

Aspect supports searches for an exact phrase or the spoken topic you remember. Once you open a candidate result, listen to the passage and use its transcript to check the wording and context.

A matching time range identifies a moment in the source. It does not by itself establish that the application creates an NLE subclip or transfers in and out points. Check the handoff supported by your editing workflow.

Check that you are searching the right project, have access to the media, and have allowed the required indexing to finish. Then inspect restrictive filters and try a simpler description of the visible or spoken content.

AI can help find a shot, but a visual match cannot establish usage rights. Filter with maintained rights information and check the underlying releases or licensing records before reuse.