Turn a stream VOD into vertical clips

Why is a stream VOD harder to clip than a talking head?

Because the interesting part is almost never in the middle of the frame.

A stream is composited before it is ever recorded. Gameplay, a shared screen or a browser fills the full 16:9 width, and the streamer sits in a small webcam box in a corner — bottom-left, bottom-right, wherever the HUD is not. That single layout decision breaks the assumption every generic clipper makes. Crop the centre and you get the middle of a game map with no face, no reaction and no idea who is talking. Track the face instead and you get a heavily upscaled corner box with the gameplay it was reacting to cut away.

The second problem is length. A talking-head video is minutes long; a VOD is hours. The moment worth clipping is four minutes somewhere inside four hours, and finding it by scrubbing is the reason most streamers' VODs are never clipped at all. That archive is the clearest case of unpublished inventory there is: hours of recorded material that reached the people already watching live and nobody else.

The third is that the audio is continuous. There are no chapter marks, no edits and no natural boundaries, so the cut points have to come from the transcript and the visual scene changes rather than from structure that is not there.

How does Clippen rebuild a stream layout?

With a mode called INSET, and it is worth explaining because the detection is the interesting part.

Clippen does not ask a language model whether the frame has a webcam box — asking made the overall layout choice worse. It finds the inset geometrically, using three tests that a corner webcam passes and a talking head does not. The subject is small in the frame. The subject is off-centre horizontally — a talking head can sit high in frame but is almost always centred, so vertical position alone is not enough. And the subject's box barely moves between samples: a webcam overlay is pinned to the same pixels, drifting a few pixels across a clip, while a real person in a room moves hundreds.

When all three hold, the clip is rebuilt rather than cropped: the screen runs full width across the top of the 9:16 frame, and the webcam box is enlarged underneath it. Both halves of the original composition survive — you can read what is on screen and see the reaction to it at a legible size.

The related SCREENCAST mode covers the other shape of the same problem: a shared screen or slide deck that spans most of the width with a presenter beside it rather than on top of it. Content narrower than half the frame is left alone, because a corner ticker or a lower-third survives a crop perfectly well and restacking it would be worse than doing nothing.

What does a four-hour VOD cost?

240 credits. Clippen bills by the length of the source, and one credit is one minute of it, so four hours is 240 minutes is 240 credits — charged once, for the whole run, however many clips come back.

Clippen is a paid tool and bills by the length of what you feed it: one credit is one minute of source video, and that one charge covers the whole run — every clip the video yields, the vertical reframing, the captions and the hook. A 60-minute podcast costs 60 credits whether it comes back with six clips or twelve. Plans start at $9.99 a month for 50 credits ($4.99 a month billed annually) and run to $199.99 a month for 1200 credits. There is no free tier.

This is worth checking carefully against any tool you compare, because per-minute billing and per-clip billing produce wildly different numbers for the same VOD. A four-hour stream that yields thirty clips costs the same 240 credits here as one that yields eight. A monthly cap on finished clips prices the same VOD by an output nobody can predict before pressing start.

The practical consequence for streamers is that trimming the source is the lever that matters. If you know the first hour was setup and chat, submit the back three hours and pay 180 instead of 240. Nothing else about the run changes.

How long does it take, and what comes back?

Processing runs at 0.30 to 0.36× real time, so a four-hour VOD takes roughly 70 to 90 minutes end to end. A 90-minute stream takes a little under half an hour. The job runs server-side: submit the VOD when the stream ends and go to bed.

Per hour of source you get 6 to 12 clips, each 15 to 60 seconds, and the count scales with what is in the material — dead air and setup do not generate clips to hit a number. Each one arrives finished:

• 1080×1920 vertical, with 9:16, 1:1 and 16:9 available from the same run.
• Word-level captions burned in with the spoken word highlighted, language detected automatically.
• An AI-written hook headline over the opening seconds.
• A layout chosen per scene: INSET for webcam-in-corner, SCREENCAST for a shared screen beside the presenter, TRACK for a full-frame facecam segment, SPLIT for a co-streamed two-shot.
• Optional punch-ins on the beats, so a fixed webcam does not read as a fixed clip.
• Download, or post to TikTok, Instagram Reels and YouTube Shorts from the dashboard.

For streams with no face on screen at all — a full-screen game capture with voice only — there is no inset to find and no subject to track, so those scenes are handled as full-width layouts with the audio and captions carrying the clip. The reframing cannot invent a reaction shot that was never recorded.