ComfyUI_ProPainter_Nodes

ComfyUI_ProPainter_Nodes
★ 373

视频修补ComfyUI节点ProPainter集成权重自动下载
在ComfyUI中实现ProPainter框架的视频修补节点集，便于逐帧修复并自动下载管理模型权重。
💡 在ComfyUI中对视频逐帧进行物体去除与修复
🍴 13 Forks💻 Python🔄 2024-12-22
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/79aaff81621b
📦 requirements.txt
opencv-python
📄 README

ProPainter Nodes for ComfyUI

ComfyUI implementation of ProPainter for video inpainting. ProPainter is a framework that utilizes flow-based propagation and spatiotemporal transformer to enable advanced video frame editing for seamless inpainting tasks.

Features

👨🏻‍🎨 Object Removal

🎨 Video Completion

Installation

ComfyUI Manager:

You can use ComfyUI Manager to install the nodes:

  • Search for ComfyUI ProPainter Nodes and author daniabib.

    • Manual Installation:

  • Clone this repository to ComfyUI/custom_nodes:

    • “`bash

    git clone https://github.com/daniabib/ComfyUI_ProPainter_Nodes

    “`

  • Install the required dependencies:

    • “`bash

    pip install -r requirements.txt

    “`

    Models will be automatically downloaded to the weights folder.

    Examples

    Basic Inpainting Workflow

    https://github.com/daniabib/ComfyUI_ProPainter_Nodes/assets/33937060/56244d09-fe89-4af2-916b-e8d903752f0d

    https://github.com/daniabib/ComfyUI_ProPainter_Nodes/blob/main/examples/propainter-inpainting-workflow.json

    Others suggested nodes

  • VideoHelperSuite for loading and saving the video frames.
  • ComfyUI-YoloWorld-EfficientSAM for masking images using prompt.

    • Nodes Reference

    🚧 Section under construction

    ProPainter Inpainting

    Input Parameters:

  • image: The video frames to be inpainted.
  • mask: The mask indicating the regions to be inpainted. Mask must have same size of video frames.
  • width: Width of the output images. (default: 640).
  • height: Height of the output images. (default: 360).
  • mask_dilates: Dilation size for the mask (default: 5).
  • flow_mask_dilates: Dilation size for the flow mask (default: 8).
  • ref_stride: Stride for reference frames (default: 10).
  • neighbor_length: Length of the neighborhood for inpainting (default: 10).
  • subvideo_length: Length of subvideos for processing (default: 80).
  • raft_iter): Number of iterations for RAFT model (default: 20).
  • fp16: Enable or disable FP16 precision (default: “enable”).

    • Output:

  • IMAGE: The inpainted video frames.
  • FLOW_MASK: The flow mask used during inpainting.
  • MASK_DILATE: The dilated mask used during inpainting.

    • ProPainter Outpainting

    Note: The authors of the paper didn’t mention the outpainting task for their framework, but there is an option for it in the original code. The results aren’t very good but I decided to implement a node for it anyway.

    Input Parameters:

  • image: The video frames to be outpainted.
  • width: Width of the video frames (default: 640).
  • height: Height of the video frames (default: 360).
  • width_scale: Scale factor for width expansion (default: 1.2).
  • height_scale: Scale factor for height expansion (default: 1.0).
  • mask_dilates: Dilation size for the mask (default: 5).
  • flow_mask_dilates: Dilation size for the flow mask (default: 8).
  • ref_stride: Stride for reference frames (default: 10).
  • neighbor_length: Length of the neighborhood for outpainting (default: 10).
  • subvideo_length: Length of subvideos for processing (default: 80).
  • raft_iter: Number of iterations for RAFT model (default: 20).
  • fp16: Enable or disable FP16 precision (default: “disable”).

    • Output:

  • IMAGE: The outpainted video frames.
  • OUTPAINT_MASK: The mask used during outpainting.
  • output_width: The width of the outpainted frames.
  • output_height: The height of the outpainted frames.

    • License

    The ProPainter models and code are licensed under NTU S-Lab License 1.0.