Comfyui-ZNGB-SceneDetect

ComfyUI custom node based on PySceneDetect. It detects scene boundaries from a loaded video, outputs one representative frame per scene, exports each scene as an mp4 segment, and returns scene metadata as JSON.

Node

• Display name: ZNGB-PySceneDetect
• Category: Video/PySceneDetect
• Class name: PySceneDetectToImages

Features

• Detect scenes with content, adaptive, or threshold mode
• Output one representative image for each detected scene
• Export each detected scene as an mp4 segment into ComfyUI temp
• Return structured scene metadata in JSON format
• Optionally write thumbnail jpg files for each scene

Installation

Clone or copy this folder into your ComfyUI custom_nodes directory:

custom_nodes/Comfyui-ZNGB-SceneDetect

Install dependencies:

pip install -r requirements.txt

Current dependencies:

• scenedetect>=0.6,<0.7
• opencv-python>=4.9,<5
• numpy

Restart ComfyUI after installation.

Inputs

Required

• image

  • Type: IMAGE
  • Usually connected from the image/frame output of a video loading node
  • LATENT input is rejected at runtime

• video_info

  • Type: VHS_VIDEOINFO
  • Connect the fourth output from VHS Load Video or another compatible video node that provides fps metadata

• method

  • Options: content, adaptive, threshold
  • Scene detection strategy used by PySceneDetect

• threshold

  • Float
  • Detection threshold for the selected method

• min_scene_len_sec

  • Float
  • Minimum scene duration in seconds

• min_scene_len_frames

  • Int
  • Minimum scene duration in frames

• luma_only

  • Boolean
  • When enabled, scene detection uses luminance only

Optional

• representative

  • Options: start, middle, end
  • Chooses which frame to output for each scene

• max_width

  • Int
  • Resize representative frames to fit within this width while keeping aspect ratio

• max_height

  • Int
  • Resize representative frames to fit within this height while keeping aspect ratio

• limit_scenes

  • Int
  • If greater than 0, only the first N detected scenes are used

• write_thumbs

  • Boolean
  • When enabled, writes thumbnail jpg files to disk

• thumbs_dir

  • String
  • Output folder for thumbnails
  • If left empty, thumbnails are written to ./scene_thumbs

Outputs

• images

  • Type: IMAGE
  • Batch of representative frames, one frame per scene

• scenes_json

  • Type: STRING
  • JSON string containing fps, method, thresholds, and per-scene metadata

• scene_count

  • Type: INT
  • Total number of scenes returned by the node

• path_string

  • Type: STRING
  • Newline-separated absolute paths of exported scene mp4 files

Exported Segment Files

Each run creates a new random output folder under ComfyUI temp:

ComfyUI/temp/transnet_segments_0833f5f6/

Inside that folder, segments are written as:

segment_001.mp4
segment_002.mp4
segment_003.mp4

This avoids filename collisions between different workflow runs.

Typical Workflow

1. Use a compatible video loader such as VHS Load Video.
2. Connect the frame/image output to image.
3. Connect the video metadata output to video_info.
4. Run ZNGB-PySceneDetect.
5. Use images for preview, selection, or downstream image workflows.
6. Use scenes_json for metadata-driven logic.
7. Use path_string if you need the exported mp4 paths in later nodes.

Notes

• The node requires a valid fps value in video_info. If fps is missing, execution will fail.
• The node converts the input image batch back into a temporary video file before scene detection.
• If no representative frame can be produced, the node returns a 1x1 black fallback image for type consistency.
• Exported scene videos are written into ComfyUI's temp directory, not the output directory.

Included Workflow

This repository includes an example workflow file:

• workflow/pysence-detect_workflow.json

License

See LICENSE.