ComfyUI-Video-Segmentation

ComfyUI-Video-Segmentation
★ 38

视频分割场景检测TransNetV2FFmpeg兼容
基于TransNetV2的自动视频场景检测与切割节点，可精确识别镜头边界并输出独立片段，用于后续逐帧处理。
💡 自动检测并切割长视频为独立镜头，便于逐段批量处理
🍴 5 Forks💻 Python🔄 2025-08-10
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/f414772aa5c3
📦 requirements.txt
numpy>=1.19.0
pillow>=8.0.0
transnetv2-pytorch
tensorflow>=2.1.0,<3.0.0
opencv-python>=4.5.0
ffmpeg-python>=0.2.0
#
TransNetV2
Model
Weights
#
The
model
weights
need
to
be
downloaded
separately
from:
#
https://huggingface.co/MiaoshouAI/transnetv2-pytorch-weights
📄 README

ComfyUI Video Segmentation Node

A ComfyUI custom node for automatic video scene segmentation using TransNetV2.

Description

This node segments videos into individual scenes automatically using the TransNetV2 deep learning model. TransNetV2 is a state-of-the-art neural network for shot boundary detection in videos.

Features

  • Automatic scene detection and segmentation
  • Uses TransNetV2 for accurate shot boundary detection
  • Fallback to OpenCV-based scene detection when TransNetV2 is unavailable
  • Configurable scene detection parameters
  • Outputs individual video segments

    • Installation

  • Clone or download this repository to your ComfyUI custom_nodes directory:

    • “`

    D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Video-Segmentation\

    “`

  • Install required dependencies:

    • “`bash

    pip install -r requirements.txt

    “`

  • Install FFmpeg (required for video processing):
  • Windows: Download from https://ffmpeg.org/download.html
  • Linux: sudo apt-get install ffmpeg
  • macOS: brew install ffmpeg

  • Download TransNetV2 model weights (for best performance):
  • Download the transnetv2-weights directory from TransNetV2 repository
  • Place it in the node directory or your preferred model directory

  • Alternative TransNetV2 installation:

    • “`bash

    pip install transnetv2

    “`

  • Restart ComfyUI

    • Usage

    Inputs

  • video: Path to the input video file (STRING)
  • output_dir: Directory where segmented videos will be saved (STRING)
  • min_scene_length: Minimum scene length in frames (INT, default: 30)
  • threshold: Scene detection threshold (FLOAT, default: 0.5)

    • Outputs

  • output_directory: Path to the directory containing segmented videos (STRING)
  • num_segments: Number of video segments created (INT)
  • segment_paths: List of paths to individual segment files (LIST)

    • Example Workflow

  • Add the “Video Segmentation” node to your ComfyUI workflow
  • Set the input video path
  • Specify the output directory (optional – defaults to ComfyUI output directory)
  • Adjust min_scene_length and threshold as needed
  • Run the workflow

    • Dependencies

    Required

  • ComfyUI
  • Python 3.8+
  • PyTorch
  • NumPy
  • PIL (Pillow)

    • Optional (for better performance)

  • TransNetV2: pip install transnetv2
  • OpenCV: pip install opencv-python (used as fallback)
  • FFmpeg (for video segment creation)

    • Technical Details

    TransNetV2 Integration

    The node first attempts to use TransNetV2 for scene detection. TransNetV2 is a neural network specifically designed for shot boundary detection and provides superior accuracy compared to traditional methods.

    Fallback Method

    If TransNetV2 is not available, the node falls back to an OpenCV-based scene detection algorithm that:

  • Analyzes frame-to-frame differences
  • Identifies sudden changes in visual content
  • Segments the video based on detected boundaries

    • Output Format

  • Segmented videos are saved as MP4 files
  • Files are named sequentially: segment_000.mp4, segment_001.mp4, etc.
  • Original video quality and codec are preserved when possible

    • Configuration

    Scene Detection Parameters

  • min_scene_length: Controls the minimum duration of detected scenes. Shorter scenes are merged with adjacent ones.
  • threshold: Controls the sensitivity of scene detection. Lower values detect more scene changes, higher values are more conservative.

    • Troubleshooting

    TransNetV2 Not Found

    If you see errors about TransNetV2 not being found:

  • Install TransNetV2: pip install transnetv2
  • Or ensure the TransNetV2 source code is available in your project

    • FFmpeg Not Found

    For video segmentation, FFmpeg is required:

  • Install FFmpeg and ensure it’s in your system PATH
  • Or install via conda: conda install ffmpeg

    • Permission Errors

    Ensure the output directory is writable and you have sufficient disk space for the segmented videos.

    References

  • TransNetV2 Paper
  • TransNetV2 GitHub Repository
  • ComfyUI Documentation

    • License

    This project is licensed under the same terms as ComfyUI. Please refer to the original TransNetV2 license for the underlying model.