Scene Detection

The scene detection node finds scene boundaries (cuts) in a video using optical-flow analysis, and emits the resulting segments. Each run replaces the whole segment set for the asset.

Kind

scene-detection

Applies to

Video only

Inputs

None

Output keys

scene_detection (String / JSON list of scene boundaries)

Requirements

OpenCV / video4j native runtime (optical-flow analysis). CPU-bound; decoding long videos is heavy.

Persists to

asset_segment_comp (whole-set replace) + asset_node_result ledger

The detector is OpticalFlowSceneDetector. Because segments are a whole-set replace, a shorter re-run deletes the surplus rows.

Configuration

No custom options beyond the common node flags.

Use Cases

  • Chaptering — split a video into navigable scenes.

  • Keyframe / per-scene thumbnails — generate a preview per scene.

  • Targeted analysis — run heavier nodes (faces, captioning) once per scene instead of per frame.