Video Metadata Search: Date, Format, Resolution

Every video file carries metadata. Some of it is in the filename and file dates. Some of it has to be read from the file: codec, resolution, colour space, frame rate. Some of it does not exist until a model generates it: transcripts, scene descriptions, object labels, face clusters.

Video metadata search uses all of these layers.

Two types of video metadata

Technical metadata

Technical metadata is written by the camera or recorder and describes the file.

Codec and container. H.264, H.265, ProRes 422 HQ, R3D, BRAW, DNxHR, XAVC.
Resolution. 1920x1080, 3840x2160, 4096x2160, 6144x3456, 8192x4320.
Frame rate. 23.976, 24, 25, 29.97, 30, 48, 60, 120.
Colour space and gamma. Rec.709, Rec.2020, S-Log3, V-Log, REDWideGamutRGB, ACES.
Duration.
File size.
Creation date. The most useful field for narrowing a search.

This metadata is present from the moment the file is written and needs no processing.

Descriptive metadata

Descriptive metadata comes from AI analysis of the content.

Transcripts. Every word spoken, with sentence timestamps and speaker labels.
Object labels. Objects detected in the frames, timestamped.
Scene descriptions. Natural-language captions of what is on screen.
Face clusters. The people who appear.
Colour characteristics. Dominant colours in the footage.

This metadata has to be generated by processing the video. Once generated it sits in the same search index as the technical fields.

How filters narrow results

Each filter removes part of the result set before you see it.

Search "product demo" across the library: 200 results
Add a date filter (last 6 months): 40 results
Add a resolution filter (4K only): 15 results
Add a codec filter (ProRes): 6 results

Date range

If you use one filter, use date. You usually know roughly when something was shot: "the March shoot," "sometime last autumn," "this year." Even a wide range removes most of the library.

FrameQuery supports absolute ranges (March 1 to March 31, 2026) and relative ranges (last 30 days, last quarter, this year). Teams that shoot often use the relative ranges for recent footage.

Resolution and codec

Resolution and codec filters find footage that meets a delivery spec.

When a client wants 4K ProRes, you need the 4K ProRes originals, not the HD proxies or the H.264 exports of the same interview. A content search cannot tell those apart. Metadata filters can.

4K or above for high-resolution deliverables
ProRes or DNxHR for an intermediate codec pipeline
R3D or BRAW for camera RAW and full grading latitude
H.264 or H.265 for compressed delivery copies

Filtering by codec across the whole library also shows the format breakdown of the archive, which helps with storage planning and transcode decisions.

Combining the two types

Technical metadata narrows the pool. Descriptive metadata finds the content.

Delivery-grade interview footage. "interview" (transcripts and scenes) + 4K + ProRes + last 3 months.

B-roll for social. "aerial city" (scene descriptions) + under 60 seconds + HD or above.

One person's footage for an archive. A face (face recognition) + 2024 date range. Every appearance of that person that year, in any format from any camera.

How FrameQuery indexes both

The initial scan reads file headers and stores technical metadata. You can filter by date, resolution, codec and duration before any AI processing runs. Indexing then generates transcripts, object labels, scene descriptions and face clusters. Both sets go into one Tantivy index and are queried together.

No manual tagging

Manual metadata entry means watching footage, typing keywords and assigning categories. It works for a few hundred clips and stops at a few thousand. AI-generated metadata covers every clip that gets indexed, and technical metadata comes from the file. Nobody has to find time to tag.

Download FrameQuery to search your footage by metadata and content.