maskilayer

maskilayer is a Python tool for compositing two images using one or more mask images.

• Composite two images (background + overlay) using one or more mask images
• Supply positive masks (overlay is preferred in bright areas of the mask) or negative masks (overlay is preferred in dark mask areas)
• Normalize masks with adjustable levels (0-5)

1. Installation

Install maskilayer using pip:

pip install --upgrade maskilayer

or the development version in a specific Python:

python3.11 -m pip install --upgrade git+https://github.com/twardoch/maskilayer

2. Rationale

2.1. Selective image sharpening

Combine a sharpened image with its original version using a segmentation mask:

1. Process an original image with an automatic sharpening method.
2. Use a segmentation model to generate a segmentation mask that isolates a specific subject.
3. Use maskilayer to composite:
   • background: original image
   • overlay: sharpened version
   • mask: segmentation mask

maskilayer will save a result, which will be a composite image where only the subject is sharpened, while the rest remains as in the original.

2.2. Creative upscaling with depth based compositing

Blend two differently upscaled versions of an image using a depth mask:

1. Upscale an image using a conservative upscaler like Codeformer to get predictable details for the background areas of the image. Supply the conservative upscale as background to maskilayer.
2. Upscale the same image using a creative upscaler like Ultimate SD Upscale to get additional details for foreground (subject) areas of the image. Supply the creative upscale as overlay (compositing image) to maskilayer.
3. Generate a depth mask using Depth Anything or Midas (where the far areas are dark, and the close areas are bright). Supply the result as the (positive) mask to maskilayer. Or use a model like Marigold (where the close subjects are dark), and supply the resulting mask as inverted (negative) mask.
4. Use maskilayer to composite:
   • background: conservative upscale
   • overlay: creative upscale
   • mask: depth mask (inverted mask if close areas are dark)

maskilayer will save a result, which will be a composite image with creative details for close subjects, and more conservative rendering for distant areas.

3. Usage

3.1. Command Line Interface

3.1.1. Basic usage

maskilayer -b background.png -c overlay.png -o output.png

3.1.2. Selective image sharpening example

maskilayer --back original.png --comp sharpened.png --out sharpened_subject.png --masks segmentation_mask.png --norm 3 --verbose

3.1.3. Creative upscaling with depth based compositing example

maskilayer --back conservative_upscale.png --comp creative_upscale.png --out composite_upscale.png --masks "depth_mask1.png;depth_mask2.png" --imasks "inverted_depth_mask3.png" --norm 2 --verbose

3.1.4. CLI documentation

NAME
    maskilayer - Composite two images using mask(s).

SYNOPSIS
    maskilayer <flags>

DESCRIPTION
    Composite two images using mask(s).

FLAGS
    -b, --back=BACK
        Type: str
        Default: ''
        layer 0 (background image path)
    -c, --comp=COMP
        Type: str
        Default: ''
        layer 1 (overlay image path that will be composited via masks)
    -o, --out=OUT
        Type: str
        Default: ''
        output composite image
    -s, --smask=SMASK
        Type: str
        Default: ''
        path to save the final mask (optional)
    -m, --masks=MASKS
        Type: Optional
        Default: None
        ;-separated mask image paths (optional)
    -i, --imasks=IMASKS
        Type: Optional
        Default: None
        ;-separated negative mask image paths (optional)
    -n, --norm=NORM
        Type: int
        Default: 0
        perform mask normalization with level 0-5
    -v, --verbose=VERBOSE
        Type: bool
        Default: False
        print additional output
    -f, --fast=FAST
        Type: bool
        Default: False
        save fast but larger files

3.2. Python API

3.2.1. Basic usage

from pathlib import Path
from maskilayer import comp_images

comp_images(
    background=Path("background.png"),
    overlay=Path("overlay.png"),
    output=Path("output.png")
)

3.2.2. Selective image sharpening example

from pathlib import Path
from maskilayer import comp_images

comp_images(
    background=Path("original.png"),
    overlay=Path("sharpened.png"),
    output=Path("sharpened_subject.png"),
    masks=[Path("segmentation_mask.png")],
    normalize_level=3,
    verbose=True
)

3.2.3. Creative upscaling with depth based compositing example

from pathlib import Path
from maskilayer import comp_images

comp_images(
    background=Path("conservative_upscale.png"),
    overlay=Path("creative_upscale.png"),
    output=Path("composite_upscale.png"),
    masks=[Path("depth_mask1.png"), Path("depth_mask2.png")],
    invert_masks=[Path("inverted_depth_mask3.png")],
    normalize_level=2,
    verbose=True
)

4. Mask handling

• If you supply multiple masks, maskilayer averages them for the final composition.
• If you supply a normalization level, maskilayer will adjust the mask contrast:
  • Level 0 uses masks as-is
  • Level 1 stretches grayscale range to full black-white spectrum
  • Levels 2-5 progressively increase contrast for more abrupt transitions between bright and dark

4.1. Tips for handling multiple mask paths

4.1.1. In CLI

• Use semicolons (;) to separate multiple mask paths (you also may use commas):

  maskilayer --masks mask1.png;mask2.png;mask3.png
  

• For inverted masks, use the --imasks flag:

  maskilayer --imasks inverted_mask1.png;inverted_mask2.png
  

• You can use both positive and negative masks in the same command:

  maskilayer --masks positive_mask.png --imasks negative_mask.png
  

4.1.2. In Python

• Use lists to provide multiple mask paths:

  masks=[Path("mask1.png"), Path("mask2.png"), Path("mask3.png")]
  

• For inverted masks, use the invert_masks parameter:

  invert_masks=[Path("inverted_mask1.png"), Path("inverted_mask2.png")]
  

• You can use both positive and negative masks in the same function call:

  comp_images(
      ...,
      masks=[Path("positive_mask.png")],
      invert_masks=[Path("negative_mask.png")]
  )
  

5. License

• Idea & Copyright (c) 2024 Adam Twardoch
• Python code written with assistance from OpenAI GPT-4o and Anthropic Claude 3
• Licensed under the Apache License 2.0