PixelOE is a Python library for generating high-quality pixel art from standard images without AI or neural networks. It targets artists and developers seeking a detail-oriented, contrast-aware approach to pixelization, offering predictable results and preserving fine details through classical image processing techniques.

How It Works

PixelOE employs a two-stage process. First, "Contrast-Aware Outline Expansion" enhances key features by generating a weight map based on local contrast and brightness, then applying selective morphological operations (erosion/dilation) blended according to these weights. This preserves important edges before downscaling. Second, "Contrast-Based Downsampling" processes luminance and color channels separately. It uses a sliding window to select the most representative pixel within each patch based on value distribution, ensuring luminance details are maintained, while color channels are smoothed with a median filter.

Quick Start & Requirements

• Install via pip: pip install pixeloe
• ComfyUI integration: Clone repo into custom_nodes directory.
• Python API: Supports legacy (NumPy/CV2) and fast PyTorch (GPU accelerated) implementations.
• Requirements: Python, PyTorch (for GPU acceleration).
• Demo: Gradio client available at ./client/demo-gr.py.
• CLI: pixeloe.pixelize and pixeloe.outline commands.

Highlighted Details

• Pure PyTorch implementation achieves >180 img/sec (bs1) on RTX4090.
• Core logic is GPU-free and can run on CPU.
• Offers multiple downscaling modes (center, contrast, k-centroid, bicubic, nearest).
• Includes optional color palette optimization and color matching.

Maintenance & Community

• Project maintained by KohakuBlueleaf.
• Mentions Claude 3 opus and Gemini 2.0 Flash for code assistance.
• Citation available in BibTeX format.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Suitable for commercial use and integration into closed-source projects.

Limitations & Caveats

The CLI is noted as using a legacy API and will be deprecated. While GPU-free operation is supported, the fastest performance is achieved with PyTorch and a GPU.