Start with footage you already know
You remember filming a wide coastal shot. You do not remember the filename, the card it came from or where it begins inside a long take. Searching filenames will only help if someone named the file well. Shot-level search lets you ask about what is visible inside it.
Metachlorian is free, open-source AI footage search. Download the desktop app, add a small, familiar folder and allow indexing and analysis to finish. The first launch downloads the analysis models. Starting small gives you a way to check whether the descriptions and search results reflect the material you actually shot.
This workflow uses example queries, not a promise that a matching shot exists in your archive. Search quality depends on the footage, completed analysis and models you have enabled.
Describe the visible moment first
Begin with a subject, an action and a setting: a person walking along a rocky coastline. Then add requirements that matter to the edit: wide, slow camera movement, at least eight seconds. A practical starting query is slow wide coastal drone shots at golden hour.
For an interview about a journey, an abstract brief such as “the feeling of getting away” may be less useful than a set of concrete searches: packing a suitcase, train leaving a station, or view through a car window. Try those separately, compare the candidates, and decide which images support the story.
If you remember something that was said, search that phrase too. Metachlorian combines visual similarity with words in transcripts, on-screen text and labels; it is useful for both silent cutaways and dialogue.
Narrow the results without hiding good options
- Start broad. Search the subject before adding every technical constraint.
- Check the parsed query. Review the editable filters and preferences rather than assuming every word became the filter you intended.
- Set the usable duration. A two-second cutaway and an eight-second establishing shot have different needs. Leave room for your final trim.
- Inspect technical details. Check resolution and frame rate against the delivery. Review the shot itself for focus, shake and composition.
- Try a similar shot. Once you find a promising candidate, use visual similarity to look for another angle or a related moment.
If a search is empty, remove a constraint and try again. Also check whether analysis is still running. “No results” does not always mean “not in the archive”.
Review the result, not just its label
Each result points to a shot within a source file. Open it, inspect the in and out points, and check the explanation for why it matched. A model label is evidence to review, not an editorial decision.
Metachlorian distinguishes measured signals, such as duration, from model output and human corrections. If a description or tag is wrong, correct it. Human corrections are stored separately and take priority after reprocessing. This makes the library more useful for the next search as well as the current edit.
Make a small collection of candidates. Keep a note about why each shot belongs in the sequence: establishing location, covering a cut, showing a detail, or changing pace. Your collection is a shortlist to edit from, not an automatically finished film.
Check recorded permissions before export
Record the source, licence, releases and relevant usage restrictions, then state the intended use when searching. For example, a paid-social edit may need different permissions from an editorial piece. Metachlorian filters against the rights recorded in the library and includes a verdict with results.
Keep those records accurate. A retrieved shot or an imported web video does not establish permission to reuse it. Downloaded web footage begins with unknown rights until someone reviews it.
Before importing private or unlisted YouTube videos: turn off Settings → Community sharing if their metadata must remain private. New YouTube imports share machine metadata with a public community index by default, separately from hosted model settings. Read the complete sharing behaviour.
Take your shortlist into the edit
Export selected shots through OpenTimelineIO, FCPXML or CMX 3600 EDL, or send a package to Cutawan. Choose a format supported by your editing workflow, then check the imported timing and media paths before building the final sequence.
If an AI agent helps with selection, it can search the same library through MCP, inspect the evidence and prepare a handoff with the required scopes. Read the MCP video search guide. For a shared archive on a server or NAS, use the self-hosted library guide.
The useful loop is simple: describe a real moment, review a few candidates, correct mistakes, and save the shots that serve your story.
Find the shot in your footage.
Download the free desktop app, or explore the illustrative sample before installing.
Download free Explore the sample demo