Clipwyn / Blog

Clipwyn guides

How Clipwyn scores a moment: two judges, one clip

How Clipwyn decides what to clip: one model reads the words, one reads the reaction, and a clip is made when both agree, scored out of 100.

By Braivex · · 5 min read

Clipwyn makes a clip when two separate judges agree that a moment is worth posting. One reads only the transcript, what was said. The other sees only the reaction: how much louder than usual the room got and, on Twitch, how busy chat was. Neither can see what the other sees. When both pick the same stretch, Clipwyn cuts it and gives it a score out of 100.

This post explains why it works that way, what the score means and what you can change.

Why one signal is not enough

A loudness meter finds every shout, including the one at a dog barking off camera. A transcript reader finds every well-formed sentence, including the ones nobody cared about. Each signal on its own produces plenty of clips and a lot of them are weak.

The moments people share have both: something worth hearing and a visible reaction to it. Asking two judges that cannot see each other's evidence is a simple way to require both. If the words judge likes a stretch and the reaction judge saw nothing there, Clipwyn leaves it.

What each judge looks at

The words judge

The whole recording is transcribed on Braivex servers, with every word timestamped. The words judge reads that transcript in order and looks for a complete moment: a setup and a payoff, a claim with its reason, a question and its answer. It is told the clip length you set (15 to 60 seconds by default), so it does not suggest something that cannot fit.

The reaction judge

The reaction judge never sees a word. It gets the stream as a series of two-second slices, each with one number for how much louder than usual the streamer was, and, on Twitch, how many people were talking in chat. It looks for a sudden jump, or a stretch that stays well above normal, and ignores music or noise that is loud the whole time.

What the score means

Every clip comes with a score out of 100, shown before you open it. A high score means both judges were confident about the same stretch. Use it to sort: open the top clips first, and treat the low ones as a second look.

The score also drives the automatic parts of Clipwyn. For live streams, a moment needs a score of 80 or more by default before it becomes a clip. For auto-posting, only clips scoring 75 or more go out on their own. You can change both numbers in a creator's settings.

Where a clip starts and ends

A clip that stops mid-sentence feels wrong even when the moment is good. Clipwyn starts each clip at the beginning of a line and ends it where a sentence ends, inside the length you allowed. If the sentence would run too long, the clip ends at the sentence before. You can still trim any clip by hand afterwards.

When the judges disagree

By default a clip needs both judges. A creator setting lets you relax that: in "either" mode, a moment that only one judge picked is still clipped if it scores 90 or more. Use it for channels where one signal is unreliable, such as a quiet streamer whose voice barely changes when something happens.

If the AI is unavailable for any reason, Clipwyn still makes clips. It falls back to the loudest stretches that contain speech, so a quiet AI day never stops the queue. Those clips have no AI hook or title until you add one.

What it does not do

The score is a guide to what to look at first, not a forecast of views. It does not know your audience, the trend of the week or what the algorithm will do. The clips that do well on your channel will teach you more than the number; the score saves you watching the whole stream to find them.

Try it on your own stream

Paste a YouTube, Twitch or Kick link, or upload a file, and Clipwyn returns scored clips with captions and a hook. Access is by approval: request access on the Clipwyn home page. For the bigger picture, Braivex has a step-by-step guide to clipping live streams for Shorts, TikTok and Reels.