Skip to main content
Scene Detect

Scene Detect · screenshot coming soon

​
What this node does

Scene Detect analyzes a COMFYTV_VIDEO, finds scene-change points, and gives you back either one thumbnail per shot or one trimmed clip per shot. You set a detection threshold and minimum gap, choose the output mode, then press ▶ Run (an ffmpeg/analysis pass on the server). The card shows the detected thumbnails, and in clips mode a scrollable pool of clips you can click to pick one. In clips mode the node also emits a picked clip based on selected_index, and its clip results are cached (up to 8 recent settings) so re-picking a different clip is fast. If no cuts are found it errors and asks you to lower the threshold.

​
When to use it

  • Break a long take or screen recording into individual shots.
  • Pull one representative frame per scene for a storyboard or contact sheet.
  • Auto-split footage into clips and pick the one you want to keep editing.

​
Parameters

​
threshold

Cut-detection sensitivity, 0.05–1.0, default 0.4. Lower = more sensitive (finds more cuts); raise it if it splits too eagerly.

​
min_gap_s

Minimum seconds between two detected cuts, 0–30, default 1.0. Prevents rapid double-cuts on flicker or fast motion.

​
output

frames (default) returns a thumbnail per shot; clips returns a trimmed clip per shot (and enables clip picking).

​
cut_mode

Only used in clips mode: fast (default) or precise. fast cuts on keyframes (quicker); precise re-encodes for exact boundaries.

​
selected_index

1-based index of the clip to emit on the clips output in clips mode. Also driven by clicking a clip in the card. Clamped to the number of clips.

​
Inputs and outputs

​
Tips

  • Start at threshold 0.4; if a hard cut is missed, lower it; if a pan/zoom triggers false cuts, raise min_gap_s.
  • Use fast cut mode for quick review; switch to precise only when frame-exact clip boundaries matter.
  • Clip results are cached per (source, threshold, gap, cut_mode) — clicking through different clips does not re-run detection.

​
Types and bridges

COMFYTV_VIDEO / COMFYTV_IMAGES are ComfyTV project snapshots, not native ComfyUI types. Use a Bridge (ComfyTV/Bridge) to interoperate with native nodes. See bridges.md.
  • Contact Sheet — arrange sampled frames into a single grid image.
  • Video Scopes — inspect a single frame’s waveform/vectorscope.