KINOPIPE FOR PODCASTERS

An episode is a long file. The work is deciding what to cut.

Transcribe the episode, get ranked clip candidates with a hook and a reason for each, cut the ones you keep, and burn the captions from the same transcript.

How an episode becomes publishable pieces
01

Send the recording to find-highlights. Files up to 2 GB, so a full video episode goes as it is. You get back candidate moments with a start, an end, a title, a hook line, the reason it works, and two scores kept apart: what happens in the clip, and how it feels to watch.

02

Read the list and keep what you want. Cut those with trim-video, then smart-crop-video for a vertical that follows whoever is speaking.

03

Run transcribe-video for the captions. Its second output is an SRT whose download URL goes straight into add-subtitles-to-video, so the words on screen come from the same transcript that chose the clip.

04

For an audio-only episode, audiogram renders the waveform over your cover art, and the same SRT burns onto it. A clip in a feed is mostly watched with the sound off.

Transcription
Parakeet, 25 languages, word timestamps
Input size for analysis
Up to 2 GB
Clip candidates
Start, end, hook, reason, two scores

Ask for the output you need.

KinoPipe gives Podcasters focused tools with bounded fields. Include the source and target format in your request.

Find the best clip candidates in this episode and show me the hooks before cutting anything.

Cut the three highest scoring moments as 9:16 with captions burned in.

Transcribe this episode and give me chapter timestamps in mm:ss.

An analysis is the only operation in its job, so the candidates come back before you spend anything rendering. You review, then you cut.

KinoPipe and Podcasters

How does it choose a moment?

The speech is transcribed on the worker with word timestamps, then a language model proposes self-contained moments and scores each one twice. Boundaries snap to what was actually said, so a clip does not open mid-word.

Does it work on audio-only shows?

Yes. The ranking comes from the transcript, so an audio file returns the same candidates. Pair it with the audiogram tool when you need something to publish as video.

Can it clean up a noisy recording?

Not today. There is no denoiser here, and a denoiser would not fix room reverb anyway, which is a different problem. Transcription runs on the audio as it is.

What does a full episode cost to analyse?

Credits track processing seconds rather than file length, and the read tools are the cheap end. The renders you choose afterwards are where the time goes.