How to generate YouTube chapter timestamps from a video
Writing chapter timestamps by hand costs about fifteen minutes an upload, and the usual shortcut still starts with extracting the audio and pasting it somewhere. Transcription that accepts the video directly removes that first step entirely, and the timestamps come back attached to the words.
The manual FFmpeg way
This is the extraction step people run before pasting the audio into a model. It is pure overhead: the file is only being converted so that something else will accept it.
ffmpeg -i episode.mp4 -vn -acodec libmp3lame audio.mp3
Do it with one API call
- 1
Send the video, not the audio
Pass the file as the input named "main". The worker pulls the audio itself, so nothing is converted on your side.
- 2
Read the timestamped segments
result.analysis holds the full text and the segments with their start and end times, in any of 25 European languages, detected automatically. result.outputs[1] is a ready-made SRT file if you also want captions.
- 3
Apply your own format
Feed the segments and your template to a model in the same run and you get the description block back already formatted, instead of reshaping it by hand on every upload.
curl https://kinopipe.com/api/v1/tools/transcribe-video \
-H "Authorization: Bearer kp_live_..." \
-H "Content-Type: application/json" \
-d '{
"inputs": [
{ "id": "main", "url": "https://example.com/episode.mp4" }
]
}'FAQ
Which model does the transcription?
Parakeet-TDT 0.6B v3, self-hosted on the worker. It detects the language among 25 European ones, punctuates, and returns word-level timestamps.
Do I have to extract the audio first?
No. Send the video file or its URL and the worker takes the audio track itself.
Can I get a caption file at the same time?
Yes. The job returns an SRT as a second output, and its download URL goes straight into the add-subtitles tool.