Chat points. You pick the highlights.

The chat-spike finder is the fastest way to get a first cut. Open a fresh VoD, click the finder, and within seconds the graph has markers on the moments chat reacted to hardest. No model is guessing what is funny; it is reading where chat got loud.

You can dial it from conservative, which only surfaces the few biggest peaks, to aggressive, which marks every meaningful jump. Most editors find a sweet spot per streamer.

The real unlock is combining the finder with the emote filter. Run it on a KEKW-only chart and you get the funny moments. Run it on a monkaS chart for the tense moments.

This is how compilation editors build a themed reel in minutes instead of hours.

Highlights appear as colored markers on the graph. Click any marker to seek to the spike, watch a few seconds of context, and decide. One click converts a marker into a draft clip with the start and end already set around the peak.

The finder is in beta because the heuristics are still being tuned. It catches the obvious peaks reliably; it can miss the quieter ones (sub stories, intimate streamer monologues that chat reacts to with text rather than emotes) and it can occasionally surface a false positive when a bot raid spams a single emote. The output is meant as a fast first pass, not a final cut.

Editors who use it heavily run the finder twice: once on the full chart for the big moments, once on a filtered chart (KEKW only, or monkaS only) to surface a single mood. That second pass tends to be cleaner because the noise floor is lower. The dial moves alongside the filter, so a conservative pass on a filtered chart returns a tight shortlist of only the strongest mood-specific peaks.

Manual scanning wins when you know the streamer, know their cadence, and the VoD is short enough (under three hours) that you can scan the whole graph in one breath. The finder wins when the VoD is long, when you do not know the streamer, or when you want a baseline shortlist you can extend rather than starting from a blank graph.

A common workflow: run the finder on conservative once for the bedrock highlights, mark the obvious keepers, then sweep the graph manually for the in-between moments it missed. Twenty minutes total instead of two hours scrubbing.