ComfyUI-ImageMotionGuider

ComfyUI-ImageMotionGuider
★ 43

图像运动潜空间引导Hunyuan Video 兼容无缝镜像
通过潜空间引导和智能镜像技术，将单张静态图生成无缝平滑的运动帧序列，便于与Hunyuan Video联合优化AI视频生成。
💡 将单张图片生成平滑连续的视频帧序列，用于AI视频生成指导。
🍴 2 Forks💻 Python🔄 2024-12-27
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/8f9eee5e2cdb
image
📄 README

ComfyUI-ImageMotionGuider

A custom ComfyUI node designed to create seamless motion effects from single images by integrating with Hunyuan Video through latent space manipulation.

https://github.com/user-attachments/assets/e6c770ed-f767-4beb-ae02-c61e9b8cfa5a

Overview

The ComfyUI-ImageMotionGuider node creates smooth motion sequences from static images through intelligent mirroring techniques. Its primary purpose is to influence AI video generation models’ behavior through latent space guidance, particularly designed for Hunyuan Video integration.

Installation

Install from Custom Node Manager

or

cd ComfyUI/custom_nodes
git clone https://github.com/ShmuelRonen/ComfyUI-ImageMotionGuider.git

Features

  • Bidirectional Motion: Horizontal movement control (-150 to 150 pixels)
  • Variable Frame Generation: Create 2 to 150 frame sequences
  • Dynamic Zoom: Optional zoom effect (0.0 to 0.5)
  • Seamless Transitions: Smart image mirroring for continuous motion
  • VAE Integration: Direct compatibility with VAE encoders
  • Hunyuan Video Ready: Designed for latent space guidance

    • Parameters

    | Parameter | Type | Range | Default | Description |

    |———–|——|——–|———|————-|

    | image | IMAGE | – | – | Input image |

    | move_range_x | INT | -150 to 150 | 0 | Motion range (positive = right, negative = left) |

    | frame_num | INT | 2 to 150 | 10 | Number of frames to generate |

    | zoom | FLOAT | 0.0 to 0.5 | 0.0 | Optional zoom intensity |

    Working with Hunyuan Video

    This node is specifically designed to enhance Hunyuan Video generation through:

  • Motion Generation:
  • Creates frame sequences with consistent directional motion
  • Maintains original image orientation
  • Ensures smooth transitions between frames

  • Latent Space Integration:
  • Compatible with VAE encoders
  • Provides motion guidance through latent space
  • Influences video generation behavior

  • Direction Control:
  • Positive values: Right-moving sequence
  • Negative values: Left-moving sequence
  • Both maintain original image orientation

    • Technical Implementation

    Motion Algorithm

  • Preserves original image orientation
  • Creates mirrored copies for seamless transitions
  • Handles directional changes smoothly
  • Combines motion with optional zoom

    • Core Features

  • PyTorch-based tensor operations
  • CUDA acceleration support
  • Efficient memory management
  • Smart edge handling

    • Usage Tips

    Image Selection

  • Use images with balanced horizontal composition
  • Avoid prominent features at edges
  • Higher resolution images recommended

    • Parameter Settings

  • Motion Range (-150 to 150):
  • Start with ±50 for testing
  • Larger values for more dramatic motion
  • Consider image content when setting

  • Frame Count (2-150):
  • 10-20 frames good for testing
  • Higher counts for smoother motion
  • Balance smoothness vs. processing time

  • Zoom (0.0-0.5):
  • Use sparingly
  • Enhances motion effect
  • Combine with motion for dynamic results

    • Hunyuan Integration

  • Workflow Setup:

    • “`

    [Image Loader] → [ImageMotionGuider] → [VAE Encoder] → [Hunyuan Video]

    “`

  • Best Practices:
  • Adjust denoise levels in Hunyuan
  • Consider frame count impact
  • Test different motion ranges

    • Requirements

  • ComfyUI
  • PyTorch
  • VAE encoder (for latent conversion)
  • Hunyuan Video (for intended usage)

    • Contributing

    We welcome contributions:

  • Bug reports
  • Feature requests
  • Pull requests
  • Usage examples

    • License

    MIT License

    Acknowledgments

  • Developed for Hunyuan Video integration
  • Built for the ComfyUI community
  • Thanks to all contributors