Skip to content

Latest commit

 

History

History
82 lines (64 loc) · 3.23 KB

ImageInterleavedUpscalerV2.md

File metadata and controls

82 lines (64 loc) · 3.23 KB

Image Interleaved Upscaler V2

A ComfyUI custom node that combines GAN-based upscaling with interlacing effects for enhanced image quality. This node applies sophisticated field-based processing to create smoother, more detailed upscaled images.

Features

  • Combines traditional upscaling with interlaced field processing
  • Supports any ComfyUI-compatible upscaling model
  • Tile-based processing for handling large images
  • Optional edge enhancement
  • Memory-efficient processing with automatic tiling adjustments

Parameters

Required Parameters

  • image: The input image to be processed
  • upscale_model: The upscaling model to use (any ComfyUI-compatible upscaler)
  • field_order: Determines the interlacing pattern
    • top_first: Process top scanlines first (default)
    • bottom_first: Process bottom scanlines first
  • blend_factor: Controls the blending between adjacent scanlines
    • Range: 0.0 to 1.0 (default: 0.25)
    • Higher values create smoother transitions between fields
    • Lower values maintain sharper field separation
  • field_strength: Controls the intensity of the interlacing effect
    • Range: 0.0 to 2.0 (default: 1.0)
    • Higher values increase the prominence of the interlacing effect
    • Lower values create more subtle field differences

Optional Parameters

  • tile_size: Size of processing tiles for large images
    • Range: 128 to 1024 (default: 512)
    • Larger values use more VRAM but may be faster
    • Automatically reduces if out of memory
  • tile_overlap: Overlap between processing tiles
    • Range: 16 to 256 (default: 32)
    • Higher values reduce visible seams but increase processing time
  • edge_enhancement: Additional edge sharpening strength
    • Range: 0.0 to 1.0 (default: 0.0)
    • Uses Sobel operator for edge detection
    • Higher values create sharper, more defined edges

How It Works

  1. The node first upscales the entire image using the provided upscaling model
  2. It then applies an interlacing effect by:
    • Separating the image into even and odd fields
    • Processing each field with the specified field order
    • Blending adjacent scanlines based on the blend factor
    • Applying field strength modulation
  3. Optional edge enhancement is applied using Sobel edge detection
  4. The final image is processed in tiles to manage memory usage

Usage Tips

  • For vintage or retro-style effects:

    • Use higher field_strength (1.5-2.0)
    • Set blend_factor lower (0.1-0.2)
    • Enable edge_enhancement (0.3-0.5)

  • For smoother, modern upscaling:

    • Use moderate field_strength (0.5-1.0)
    • Set blend_factor higher (0.3-0.5)
    • Keep edge_enhancement low or disabled

  • For large images:

    • Adjust tile_size based on your GPU VRAM
    • Increase tile_overlap if you notice seams
    • The node will automatically adjust if it runs out of memory

Memory Management

The node includes automatic memory management features:

  • Dynamically adjusts tile size if out of memory
  • Cleans up GPU memory after processing
  • Moves models between CPU and GPU as needed

Output

Returns a single upscaled image tensor in BHWC format (Batch, Height, Width, Channels), compatible with other ComfyUI nodes in the image processing pipeline.