KinoPipe
Start for free
All toolsfind highlights in long video API

Find Highlight Clips in a Long Recording

Turn a talk, podcast, stream or interview into ranked clip candidates with start, end, hook, title and a reason, so an agent can cut shorts from hours of footage without watching it. Run the preset directly on this page or call the same agent-ready endpoint from your product.

Run this tool View API recipe
LIVE TOOL
Run find highlights
Drop a video, adjust the defaults if you need to, and download the result here.

or ·

Already have an account? Sign in

Useful defaults, typed options.

The tool slug stays stable while your agent supplies named media inputs and a narrow set of documented options.

  • Recordings up to 2 GB, hours long
  • Two independent scores plus a hook and a title per clip
  • Timestamps ready for trim-video and format-for-platform
What this preset does
  1. 01Transcribe the speech on the worker (Parakeet, 25 languages)
  2. 02Ask a language model for self-contained moments that open on a hook, within your duration window
  3. 03Score each moment twice: what happens, and how it feels to watch
  4. 04Snap boundaries to sentence starts and ends, drop overlaps, return the best first as JSON
Stable endpointPOST /api/v1/tools/find-highlights
Live API request
{
  "inputs": [{ "id": "main", "url": "https://example.com/podcast-episode.mp4" }],
  "options": { "min_duration": 20, "max_duration": 60, "max_highlights": 6, "focus": "practical advice" }
}
Successful output example
{
  "result": {
    "output": {
      "filename": "find-highlights-output.mp4",
      "contentType": "video/mp4",
      "byteSize": 437021,
      "downloadUrl": "https://cdn.kinopipe.com/…"
    }
  }
}

Know the boundaries before you run.

Highlights are chosen from what is said: silent footage, music or gameplay without commentary returns no_speech_detected.
Speech in languages outside the 25 supported ones produces unusable candidates.
Worker time scales with the recording: about a tenth of the duration for transcription, so a two-hour recording takes around fifteen minutes.

About find highlights

result.analysis.highlights, best first: rank, start, end, duration, score (0-100), contentScore, viewerScore, hook (the first sentence heard), reason, title and the transcript of the clip, plus the SHA-256 of the source so you can cache the result.

Two independent scores per clip: content, for how significant or informative the moment is on its own, and viewer, for how funny, tense or satisfying it is to watch. The combined score is their geometric mean with a bonus when both are high, which favours moments that are both substantial and enjoyable.

Call trim-video with its start and end, then format-for-platform for 9:16, and add-subtitles-to-video with the SRT from transcribe-video. An agent can run that for every highlight in the list.

Related media tools

Transcribe Video to Text and SRT

Turn the speech in a video or audio file into timestamped text and a ready-to-burn SRT caption file, with an open model that runs on our own workers rather than a third-party speech API.

Explore

Trim Video with an API

Cut a precise segment from a source video using start and end times.

Explore

Create a Social Video Clip

Cut and reframe a source video for Shorts, Reels or vertical social feeds.

Explore