This project does not embed or port the full OpenToonz application. It extracts the useful production ideas for the current sketch-to-color research pipeline:

scan-style blue line cleanup,
gap closing before region fill,
region/palette maps for later color correction.

The intended upstream/downstream flow is:

blue rough panel crop
  -> OT Blue Line Cleanup
  -> OT Line AutoClose
  -> OT Region Palette Map
  -> Qwen Image Edit / MangaNinja / reference colorization
  -> panel reassemble

Why These Features

OpenToonz is strongest in traditional 2D animation cleanup, ink/paint, palette management, and compositing. For this research, the reusable ideas are not the UI or timeline; they are the low-level line-preparation steps that make rough art more fillable and controllable.

Relevant OpenToonz references:

Nodes

OT Blue Line Cleanup

Classifies blue rough pencil strokes and emits a normalized black-on-white line image plus an overlay preview.

Outputs:

clean_line: white background with black extracted lines
line_overlay: source image with detected strokes tinted red
transparent_line_preview: line-only preview on white
settings_json: exact parameters used

OT Line AutoClose

Finds skeleton endpoints and draws short closure segments when endpoint distance and direction are compatible.

Outputs:

closed_line: line image after gap closure
closure_overlay: original black lines plus red closure strokes
segments_json: closure segment coordinates

OT Region Palette Map

Labels fillable regions separated by line art and returns a JSON list of region metadata. This is the first step toward OpenToonz-style indexed color editing.

Outputs:

region_preview: colored region-label preview with black line art
regions_json: bbox, area, centroid, and style id for each region

Install

Clone or copy this folder into ComfyUI custom nodes.

On another ComfyUI installation:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/l2dnjsrud/ComfyUI-OpenToonzLineTools.git

Because this repository is currently private, authenticate first on machines that do not already have GitHub credentials:

gh auth login
gh repo clone l2dnjsrud/ComfyUI-OpenToonzLineTools /path/to/ComfyUI/custom_nodes/ComfyUI-OpenToonzLineTools

On this Mac, using the development copy by symlink is fine:

cd /Users/iwongyeong/AI/ComfyUI/custom_nodes
ln -s /Users/iwongyeong/AI/ComfyUI-OpenToonzLineTools ComfyUI-OpenToonzLineTools

Install dependencies in the ComfyUI environment if needed:

cd /path/to/ComfyUI
source .venv/bin/activate
pip install -r custom_nodes/ComfyUI-OpenToonzLineTools/requirements.txt
deactivate

Do not reinstall or upgrade torch from this node. Use the torch build that already belongs to that ComfyUI installation, especially on CUDA or MPS setups.

Restart ComfyUI. Nodes appear under:

manga/opentoonz-line-tools

Example Workflows

Use the UI workflow when opening a graph in the ComfyUI canvas:

examples/opentoonz_line_tools_basic_ui.json

This graph is:

LoadImage
  -> OT Blue Line Cleanup
  -> OT Line AutoClose
  -> OT Region Palette Map
  -> SaveImage debug outputs

Before queuing it, put a rough panel image into the ComfyUI input folder and select it in the LoadImage node. The placeholder filename is panel_001.png.

Use the API prompt version only for /prompt-style scripted runs:

examples/opentoonz_line_tools_basic_api.json

Verification

Run the smoke tests with the ComfyUI Python environment:

cd /path/to/ComfyUI/custom_nodes/ComfyUI-OpenToonzLineTools
/path/to/ComfyUI/.venv/bin/python -m unittest discover -s tests -v

Current Scope

Implemented:

OpenToonz-inspired blue cleanup baseline
endpoint-based line autoclose
fillable region labeling and palette-preview JSON
ComfyUI node wrappers
English/Korean docs and visual pipeline notes

Not implemented yet:

direct OpenToonz C++ binding
TLV/PLI file support
full palette-index raster format
semantic inking or AI colorization models

Recommended Next Experiment

Use the current Mr.Blue v5 polygon crops as input and compare:

raw crop -> Qwen Image Edit,
cleanup crop -> Qwen Image Edit,
cleanup + autoclose + region map -> reference colorization/correction.

Success should be measured by blue-line removal, composition retention, fill leak reduction, and whether palette/region JSON makes post-correction easier.