waifu-scorer

A deep learning-based tool for scoring anime-style images, supporting multiple hardware backends and PyTorch environments.

Installation

You need Python 3.10+ and pip. It is recommended to use a virtual environment.

1. Install PyTorch for your platform

Pick the right command for your OS / hardware from the official selector: https://pytorch.org/get-started/locally/

A few common examples:

# CPU only (any OS)
pip install torch --index-url https://download.pytorch.org/whl/cpu

# NVIDIA GPU (pick the CUDA version that matches your driver)
pip install torch --index-url https://download.pytorch.org/whl/cu124
pip install torch --index-url https://download.pytorch.org/whl/cu128

# AMD GPU on Linux (ROCm)
pip install torch --index-url https://download.pytorch.org/whl/rocm6.2

# macOS (Apple Silicon, MPS backend) — just use the default PyPI wheel
pip install torch

2. Install waifu-scorer

pip install waifu-scorer

If you are using uv, you can let it pick a torch backend automatically:

uv pip install --torch-backend=auto waifu-scorer

Usage in Python

You can also use waifu-scorer directly in your Python code:

from waifu_scorer.predict import WaifuScorer

scorer = WaifuScorer()
results = scorer(["path/to/image1.jpg", "path/to/image2.png"])
for img_path, score in zip(["path/to/image1.jpg", "path/to/image2.png"], results, strict=False):
    print(f"{img_path}: {score:.3f}")

Usage from Command Line

After installation, you can use the command line interface to score images:

python -m waifu_scorer path/to/image1.jpg path/to/image2.png

Options

• --model: Path to a custom model file
• --device: Device to use
• --verbose: Enable verbose output

Example:

python -m waifu_scorer examples/waifu1.png --verbose

Reference

This project refers to waifuset.

For more details, see the code and documentation in the repository.