Using PySceneDetect to cut live baseball broadcasts into highlight clips at the 2026 Asian Games

[sungbin1015]

We use PySceneDetect in SceneMaker, a system that turns long broadcast video into short clips. We ran it with SBS, a Korean broadcaster, on Team Korea's baseball games at the 2026 Asian Games. It drafted 79 clips, operators reviewed them, and they went out on SBS's "AI PICK" page.

PySceneDetect is the first step. ContentDetector with threshold=15 and min_scene_len of about 1 second splits the broadcast into shots, then each shot gets one representative frame for a vision model to describe.

The main thing we learned is that a cut isn't an event. One at-bat can span many camera changes, so we merge neighbouring shots using the commentary and the scoreboard (inning, outs, count, score) before searching them for clips. We also cap long scenes, because fades and dissolves can slip past the detector and leave one very long shot.

Clips: https://programs.sbs.co.kr/sports/ag2026/clips/90399
Docs: https://doc.scenemaker.solbox.com

Built by Solbox and SBSi under the 2026 NIPA Open-Source AI/SW Development and Utilization Support Program.

Has anyone compared AdaptiveDetector and ContentDetector on sports broadcasts with replay transitions? We've only run ContentDetector in production, and those transitions are where I'd expect the most difference.