patcolour

Partial color processing for visual-novel style artwork.

patcolour keeps selected regions in color while the rest becomes monochrome. It is meant to sit between raw photo editing and the larger skirts-colour production pipeline, and to be callable from other tools that need stable image-processing behavior.

Why patcolour?

The target use case is not generic "selective color" photography. The real goal is more specific: keep emotionally important parts in color while the rest falls back to memory-like grayscale. That makes mask quality, edge softness, and region selection workflow more important than a huge option surface.

Just as importantly, this tool is expected to become reusable infrastructure for other repos, including name-name-based skirts-colour workflows and future image-processing utilities.

Install

uv tool install patcolour

Or from source:

git clone https://github.com/kako-jun/patcolour.git
cd patcolour
uv sync --group dev
uv run patcolour --help

Usage

Mask-based

patcolour photo.jpg --mask mask.png -o output.png

White areas in the mask stay in color. Black areas become monochrome.

Coordinate-based

patcolour photo.jpg \
  --rect 120,40,220,180 \
  --ellipse 540,260,110,80 \
  --feather 12 \
  -o output.png

Auto-detect colorful regions

patcolour photo.jpg --auto-detect -o output.png

Sample a target color from the image

patcolour photo.jpg --sample 420,180 --lab-radius 16 -o output.png

This mode samples one pixel from the input and keeps nearby colors using Lab chroma distance, which is a better fit than raw RGB distance for "keep this hue family" workflows.

Keep only the front flower, not every similar color in the scene

patcolour photo.jpg \
  --sample 420,180 \
  --lab-radius 16 \
  --ellipse 420,180,90,120 \
  --feather 10 \
  -o output.png

Color alone is not enough for many scenes. In practice, the user often needs to combine a color sample with a spatial guide.

Batch processing

patcolour ./photos/ --mask-dir ./masks/ -o ./output/

Current behavior

• --mask uses an external grayscale mask
• --rect and --ellipse define keep-color regions
• --auto-detect keeps broad HSV-detected colorful areas
• --sample keeps colors near a sampled reference point in Lab chroma space
• --feather softens coordinate-based region edges

The CLI is the first interface, but not the only intended interface. The filter logic should stay predictable enough to be reused from other tools.

Just as importantly, this is a human-guided tool. If the scene contains several similar purples, reds, or greens with different semantic roles, the user must be able to tell the tool which region is the intended subject.

If --auto-detect is combined with coordinate regions, the result is their intersection.

Limitations

• Auto-detection is currently tuned for broad colorful regions and is easy to over-select
• There is no semantic selection yet such as "left flower only" or "girl's ribbon only"
• Batch mode currently assumes one mask per file and does not validate coverage quality

Docs

• docs/overview.md — positioning, goals, non-goals
• docs/spec.md — CLI and processing behavior
• docs/roadmap.md — internal roadmap and quality targets
• CLAUDE.md — AI-facing internal guide

Development

uv sync --group dev
uv run ruff check .
uv run pytest