KinoPipe
Start for free
API GUIDE

How to turn a long podcast into short clips with an API

Cutting an episode into clips is rarely slow because of the cutting. It is slow because of the deciding: which moment carries on its own, what the hook is, whether it suits a Short or a Reel. This is how to get that decision back as data you can review in a minute, then cut only what you kept.

Try find highlight clipsRead the docs

The manual FFmpeg way

FFmpeg cuts a clip once you already know the timestamps. Finding them is the part it cannot help with, which is why most of the hour goes into scrubbing rather than encoding.

ffmpeg -i episode.mp4 -ss 00:14:32 -to 00:15:20 -c copy clip.mp4

Do it with one API call

  1. 1

    Send the episode to find-highlights

    Pass the recording as the input named "main". Inputs up to 2 GB are accepted, so a full video episode is fine as is.

  2. 2

    Review the candidates

    result.analysis.clips returns start, end, title, hook, reason, and two separate scores: content for what happens, viewer for how it feels to watch. Reading a short list with one sentence each takes about two minutes.

  3. 3

    Cut the ones you keep

    Send each start and end to trim-video, then format-video-for-platform for a Short, a Reel or a TikTok. The per-platform choice is a second pass on what survived, not a separate hunt.

REST
curl https://kinopipe.com/api/v1/tools/find-highlights \
  -H "Authorization: Bearer kp_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": [
      { "id": "main", "url": "https://example.com/episode.mp4" }
    ]
  }'
Or just ask your agent
Connect the KinoPipe MCP server and say it in plain language:

“Find the best clip candidates in episode.mp4, then cut the top three as vertical videos with captions.”

Connect MCP

FAQ

The speech is transcribed on the worker with word timestamps, then a language model proposes self-contained moments and scores each one twice: once for what happens, once for how it feels to watch. Boundaries snap to what was actually said, so a clip does not start mid-word.

No. An analysis is the only operation in its job, so it returns the candidates and you cut the ones you keep with trim-video afterwards. That is deliberate: you review before you spend credits rendering.

Yes. The ranking comes from the transcript, so an audio file returns the same clip candidates. Pair it with the audiogram tool if you want something to publish as video.
Open the find highlight clips tool ← All guides