Turn one long video into the clips worth posting
Drop in a YouTube, Twitch or Kick link, or upload the file. A vision model watches the whole recording, finds the moments with a real setup and a real payoff, and returns them as vertical clips with captions burned in. No timeline, no scrubbing, no manual export.
It is built for the case where the long video is genuinely long and you have no intention of watching it back to find the good parts.
The three steps
One: give it the video. Paste a link to a YouTube, Twitch or Kick VOD, or upload your own file. You can add a prompt describing the kind of moment you are after, or leave it to decide.
Two: it watches. A self-hosted vision model samples the frames at 1.4 fps alongside the audio and transcript, building an account of what actually happens rather than a list of what was said. This is the part that takes time, roughly the runtime of the video plus overhead.
Three: you get clips. Vertical, captioned, sized for TikTok, YouTube Shorts and Instagram Reels, with the facecam auto-cropped if there is one. Download the ones you want.
Why the choosing matters more than the cutting
Cutting a video into vertical segments is a solved problem and every tool does it competently. What varies enormously is which segments you are handed, and that comes down to what the tool looked at to decide.
Most tools read the transcript and pick from the text. On a two-hour stream, that means every moment where the interesting thing was visual is invisible to the selection. The reaction, the near miss, the thing going wrong in the background: none of it is in the transcript, so none of it can be ranked.
Watching the frames is more expensive and slower. It is worth it precisely when the video is long and unscripted, because that is where the ratio of visual moments to spoken ones is highest, and where scrubbing it yourself would cost you the most time.
How many clips come out of a long video
It depends on the video, and deliberately so. You pay for the minutes you put in rather than the clips you get back, so there is no quota to spend down and no reason for the tool to pad the output.
A ten-minute video typically returns around three. Longer recordings give up proportionally more. When a stretch of footage genuinely has nothing worth posting in it, you get fewer clips rather than filler, which is the behaviour you want from something whose entire job is judgement.
Questions
- How long can the source video be?
- Long recordings are the case this is built for. What you are billed on is the length of the source video: the free tier covers 60 minutes a month, Starter $18 covers 180 minutes, Creator $30 covers 330 and Studio $60 covers 720 or more. A single long VOD simply draws against that allowance.
- Do I have to watch the video first or mark timestamps?
- No. That is the entire point. You can optionally add a prompt describing the kind of moment you want if you have something specific in mind, but the default is to hand over the video and let the model decide.
- What format do the clips come back in?
- Vertical 9:16 with auto-captions burned in, ready for TikTok, YouTube Shorts and Instagram Reels. If the footage has a facecam, the AI webcam reframe crops it for vertical automatically rather than leaving you to reposition it.
- How long does the job take?
- Roughly the runtime of the source video plus overhead. A long recording therefore takes a while, because the model genuinely watches all of it. Queue it and get on with something else. If you need clips back in minutes, a transcript-only tool will be faster and we would rather tell you that than waste your time.
- How long do I have to download them?
- Clips stay available for 14 days from creation on every tier, including free. Download anything you want to keep.
See what it finds in your footage
The free tier gives you 60 minutes of source video a month. Paste a YouTube, Twitch or Kick link, or upload your own file.
No card required. See pricing for what the paid tiers add.