ComfyUI_video_essentials

ComfyUI Video Essentials

A collection of essential custom nodes for ComfyUI, focused on video processing, batch management, and high-quality detail restoration.

Features

• Batch Management: Tools to split, trim, and combine image batches.
• Video Helpers: Utilities for looping and transitions.
• Detail Restoration: Advanced frequency-based restoration to fix texture degradation in video generation (e.g., from SVD or WAN models).
• Optical Flow: Advection and warping tools.

Installation

1. Navigate to your ComfyUI custom_nodes directory.
2. Clone this repository:

   git clone https://github.com/yourusername/ComfyUI_video_essentials.git
   

3. Install dependencies (if not already present):

   pip install -r requirements.txt
   

   (Note: This pack relies on standard libraries like torch, torchvision, and opencv-python)

Node Reference

| Node Name | Description | | :--- | :--- | | Get Images (Front/Back) | Retrieves the first N and last N images from a batch. Useful for creating loops or checking start/end frames. | | Trim Image Batch | Removes frames from the start or end of a batch. | | Image Batch Splitter | Splits a large batch into smaller chunks (e.g., for processing in parts to save VRAM). | | Image Batch Multi (Crossfade) | Combines two image batches with an optional crossfade transition. | | Optical Flow Advection (EbSynth) | Uses RAFT optical flow to advect (push) pixels from a source frame to target frames. | | Procedural Detail Restoration V2 | [Beta] Advanced restoration using Laplacian Pyramids and Optical Flow. Restores lost texture details while masking out occluded areas to prevent ghosting. | | Procedural Detail Restoration | (Legacy) Simple frequency-based restoration. |

Spotlight: Procedural Detail Restoration V2

This node is designed to fix the "waxy" or blurry look often seen in AI video generation. It works by:

1. Decomposition: Breaking the image into frequency bands (Laplacian Pyramid).
2. Alignment: Aligning the reference image to the current frame using Optical Flow (RAFT).
3. Injection: Injecting high-frequency details from the reference into the current frame, but only where they align correctly.

Usage Tips:

• Occlusion Threshold: Adjusts sensitivity to changes. Lower values are safer (less ghosting), higher values restore more detail.
• Mask Dilation: Expands the safety mask around moving objects to prevent halos.

Contributing

This project uses a modular structure. Each node is contained in its own file within video_essentials/nodes/.