Search inside video

Search Livestream Recordings by Visuals or Spoken Words

Stop relying on filenames and memory. Obskura can search what appeared on screen and what someone said across one recording, one creator folder, or your full library.

Try free for 3 days

Visual search

Describe what you remember in normal words and search timestamped frame embeddings for visually related moments.

Speech search

Use local Whisper transcription and full-text search to find a word or phrase someone said.

Search one file or the whole library

Limit a search to the current recording, a creator's folder, or every analyzed recording.

Turn results into clips

Preview a timestamp, save useful results as clip candidates, refine the boundaries, and export the final cut.

Find a visual moment from a plain-language description

Visual search samples frames and stores CLIP embeddings locally. A search such as “red dress dancing,” “scoreboard,” or “high energy moment” compares your words with those frames and returns timestamped matches with thumbnails.

Open a result to preview the recording at that timestamp. The embeddings can be reused for future searches, so the same recording does not need to be fully processed from the beginning every time.

Search for something someone said

Speech search transcribes audio locally with Whisper and stores timestamped transcript segments in SQLite full-text search. You can search exact phrases, keywords, prefixes, and supported boolean combinations.

Click a transcript result to seek to that section of the video. A useful phrase can become a saved clip candidate instead of remaining a one-time search result.

Use activity and scene changes when you do not know what to search

Sometimes you remember that something happened but not what it looked like or what was said. Obskura can analyze audio level, silence, motion, and scene changes to build an activity timeline and suggest potentially useful ranges.

Candidate boundaries can snap to nearby scene changes and include pre-roll or post-roll. The explanation view shows which signals contributed to a suggested clip.

Keep recording content private

Visual embeddings, transcripts, activity signals, searches, and clip synthesis run on the recording computer. The workflow does not require uploading the video library to an external search service.

Once you find the timestamp, open the livestream video editor to trim the result or combine it with other moments.

Find the moment instead of scrubbing for it

Record locally, analyze locally, and jump directly to the timestamps that match your search.

Try free for 3 days