Import and understand footage

Footage becomes useful for semantic search in stages. Import adds a video to the library and starts local preparation, including shot detection. AI understanding then creates the information used to describe and search those shots. A completed import is not the same as completed AI processing.

Choose the source and collection

Keep the original video in a stable location. If it is on an external drive, connect that drive before starting. Pick the collection that should contain the footage, or create one with a clear project name.

For a first batch, use a few files with recognizable content. This gives you a straightforward way to check the result before processing a larger archive. Check your account allowance and the cloud processing description before submitting material for understanding.

Import videos or a folder

1. Choose Import Video or Import Folder.
2. Select the files or folder you want to add.
3. Choose the intended target collection.
4. Start importing and allow the local preparation tasks to finish.

Shot detection divides a video into detected shots. You can inspect the resulting video and shot information in the library. The detected boundaries are an automatic interpretation of the footage; a long take or a subtle cut may not divide exactly where you would make an editorial selection.

Run AI understanding

Open the collection, select the shots you want to process, and choose AI understanding. Wait for processing and indexing to complete. Use the app's completion information and searchable-shot count to determine whether the material is ready.

Descriptions and search representations are separate results of AI processing. A video may already be present and have detected shots while its understanding tasks are still running. Do not treat a generic ready or imported state as proof that semantic search is available for every shot.

Check a known example

After completion, search for something visible in the imported video, such as “a red cup on a table.” Open the returned shot and check its actual content and range. The description may summarize only part of a shot, and a retrieved shot may include extra material before or after the moment you need.

An expected result is searchable footage with a source reference you can review. It is not a guarantee that every cut, action, or generated tag is exact. Finish precise trimming in your editing application when necessary.

If the material is not searchable

Check whether understanding is still queued or running, whether the account is signed in, and whether the footage is inside the collection being searched. If an error appears, record it before retrying.

Repeatedly importing the same file is not a reliable way to restart understanding. For a video already in the library, use its existing shots and the app's AI understanding flow. Avoid moving the original as a troubleshooting step: exports depend on its path.

For a search that completes but returns poor candidates, continue with Search your library. For stalled tasks or source-access errors, use Troubleshooting.