Magic Gadgets included!

Why use this fork? It's like having a 4th-dimensional pocket for your manga.

Folder Batch Mode

Process entire chapters at once! Just point to a folder and let it run. No more one-by-one drudgery.

Big Light (4x Scaling)

Includes Real-ESRGAN to enlarge your pages 4x! Turn blurry low-res scans into crisp HD masterpieces.

Translation Bread

Okay, it doesn't translate, but colorization makes reading raw Japanese manga much easier for learners!

3 Steps to Adventure

1

Load Images

Select a file or a folder.

2

Configure

Toggle Upscale or Denoise!

3

Start!

Watch the magic happen.

Control Panel

Master the settings to balance speed and quality.

Basic Options

Processing Toggles

Enable Colorize: The main event. Applies AI colorization.
Enable Upscale: Uses Real-ESRGAN to increase resolution by 4x.
Enable Denoise: Smooths out compression artifacts before coloring.

Preview Mode

Before running a full batch, click Preview to test settings on a random image from your input folder. Use Repick to try another random page.

Advanced Settings

Upscaler Tile Size

Default: 256

Processes the image in chunks to save VRAM.
Lower this if you crash with "Out of Memory". Set to 0 to disable tiling (requires massive VRAM).

Colorized Image Size

Default: 576

The target width for the colorization AI.
576px is the sweet spot. Higher values (e.g., 1024) might look sharper but can introduce weird artifacts.

Denoise Sigma

Default: 25

Strength of the denoising filter. Increase for very noisy/grainy scans, decrease for high-quality clean raws.

Local Setup

Follow these steps to build your own Manga factory.

1 Environment Setup

git clone https://github.com/sepTN/Manga-Colorizer-GUI cd Manga-Colorizer-GUI # Recommended: Create a clean Conda environment conda create --name manga-colorizer python=3.9 conda activate manga-colorizer

2 Install Dependencies

// 1. Install App Requirements

pip install Flask Flask_Cors matplotlib numpy opencv_python_headless scikit_image einops

// 2. Install PyTorch (Select your OS)

Mac (M1/M2/M3)

pip install torch torchvision

Windows/Linux (NVIDIA)

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

3 Download Models

Required Model Weights

You need the generator weights to run the colorization. Extract the downloaded zip into the Backend/networks/ folder.

Download Weights

Troubleshooting

ModuleNotFoundError

Usually means you forgot to activate your environment or missed a dependency.
Run: conda activate manga-colorizer

Slow Performance

Ensure PyTorch is using your GPU.
If on Nvidia, check `nvidia-smi`. If on Mac, ensure you installed the default `torch` package which supports MPS.