Skip to content

Latest commit

 

History

History
82 lines (63 loc) · 3.05 KB

VideoInterlaceGANV3.md

File metadata and controls

82 lines (63 loc) · 3.05 KB

VideoInterlaceGAN V3

A ComfyUI custom node for high-quality video frame upscaling that combines interlaced processing with GAN-based upscaling models. This node provides advanced temporal compensation and edge enhancement features for superior video quality.

Features

  • GAN-based upscaling with external model support
  • Interlaced field processing with customizable field order
  • Temporal compensation for improved frame consistency
  • Edge enhancement capabilities
  • Tiled processing for handling large frames
  • Memory-efficient operation with automatic OOM handling

Parameters

Required Parameters

  • images: Input video frames in tensor format (IMAGE)
  • upscale_model: External upscaling model to use for processing (UPSCALE_MODEL)
  • field_order: Field interlacing order
    • top_first: Process top field first (default)
    • bottom_first: Process bottom field first
  • blend_factor: Controls the strength of field blending
    • Range: 0.0 to 1.0 (default: 0.25)
    • Higher values create smoother transitions between fields
  • temporal_radius: Number of neighboring frames to consider for temporal compensation
    • Range: 1 to 3 (default: 1)
    • Higher values provide more temporal consistency but may reduce detail

Optional Parameters

  • tile_size: Size of processing tiles for memory efficiency
    • Range: 128 to 1024 (default: 512)
    • Smaller values use less memory but may be slower
  • tile_overlap: Overlap between processing tiles to prevent artifacts
    • Range: 16 to 256 (default: 32)
    • Higher values reduce tile boundary artifacts
  • enhance_edges: Strength of edge enhancement
    • Range: 0.0 to 1.0 (default: 0.0)
    • Higher values increase perceived sharpness

Technical Details

Processing Pipeline

  1. Temporal Compensation

    • Analyzes neighboring frames within the specified temporal radius
    • Applies weighted averaging for consistent motion
    • Handles edge cases at sequence boundaries

  2. Interlaced Processing

    • Upscales full frame using the provided GAN model
    • Separates fields based on specified field order
    • Applies field blending for smoother transitions

  3. Edge Enhancement

    • Uses Sobel operator for edge detection
    • Applies controllable edge enhancement
    • Preserves natural image appearance

  4. Memory Management

    • Implements tiled processing for large frames
    • Automatically handles out-of-memory situations
    • Efficiently manages GPU resources

Usage Tips

  1. Optimal Field Order

    • Use top_first for most modern video sources
    • Try bottom_first if you notice field artifacts

  2. Memory Optimization

    • Reduce tile_size if encountering memory issues
    • Increase tile_overlap if seeing tile boundary artifacts

  3. Quality Tuning

    • Adjust blend_factor for optimal field transition smoothness
    • Use enhance_edges sparingly to avoid over-sharpening
    • Increase temporal_radius for smoother motion at the cost of processing time

Category

image/upscaling