Search evidence
A real shot search test: what “close-up” returned from 163 analyzed shots
Updated September 6, 2026 · 6 min read
This is a recorded product test, not a claim that every result is correct. We searched 163 analyzed shots from Charge and Cosmos Laundromat: First Cycle with the single query “close-up,” then reviewed what appeared and where human judgment was still needed.
The test setup
The library in this recording contains two Blender open movies: Charge and Cosmos Laundromat: First Cycle. ShotMind split the local source videos into shots, and 163 selected shot clips had already been analyzed before this search.
The recorded interface reports 50 results shown out of about 85 candidates for the query “close-up.” That number is a retrieval pool, not a claim that all 85 shots are editorially correct close-ups.
- Source videos: two openly licensed Blender films.
- Search scope: 163 previously analyzed shots.
- Actual query: close-up.
- Recorded output: 50 shown out of about 85 candidates.
What matched well
Several high-ranked results are visually consistent with the query: character faces fill much of the frame, a hand and control surface are shown at close range, and small objects or textures occupy most of the image.
These results are useful because the filenames only identify the source films. They do not say that a particular moment contains a face, hand, device, or detailed texture.
Where the results were too broad
The same result set also includes a title card, wider character views, a landscape-like frame, and shots where the subject is prominent but not a strict cinematography close-up. Those are reasonable semantic neighbors, but they would not all survive an editor's final review.
Similarity is not the same as a final edit decision. ShotMind helps narrow a library to candidates; it does not replace opening the result, checking its time range, and reviewing the original footage in context.
How to use this result in a real workflow
Use a remembered visual description to reduce the search space, scan the ranked shots, and then open the promising results before choosing an edit or reference. If a broad term returns too much, add a subject, action, lighting condition, or camera detail to make the next query more specific.
This case demonstrates why shot-level search can complement folders and whole-file organization. The source files remain local, while the searchable unit is the visual moment inside each video.
Sources, license, and endorsement boundary
Charge and Cosmos Laundromat: First Cycle are Blender open movies available under CC BY 4.0. They are used here only to demonstrate ShotMind's import, shot analysis, and search workflow.
Blender, Blender Studio, Blender Foundation, and the original film teams do not endorse ShotMind. The source pages below provide the film and license records used for this case.