waifu2x-ncnn-vulkan is an image upscaling and denoising tool that runs on Intel, AMD, NVIDIA, and Apple Silicon GPUs using the Vulkan API.
The tool solves the problem of upscaling and denoising images, particularly anime-style artwork, while maintaining quality. It uses the ncnn neural network inference framework to run waifu2x models directly on GPU hardware. The approach leverages Vulkan as a universal graphics API to achieve fast processing across different GPU manufacturers without requiring vendor-specific runtimes like CUDA or proprietary frameworks.
The tool is suitable for anyone needing to upscale images on consumer hardware without installing heavy dependencies. It works with both individual files and batch processing of directories. Users can control noise reduction intensity, upscaling factor, tile size for memory-constrained GPUs, and thread counts for different processing stages. The tool supports output in PNG or WebP formats. It is portable and self-contained, bundling all necessary binaries and neural network models in downloadable releases for Windows, Linux, and macOS. The README does not name alternatives for direct comparison.
The project maintains active continuous integration workflows. Build support spans multiple platforms with platform-specific considerations, including optional static linking of the Vulkan loader on macOS. The codebase is written in C++ and integrates the ncnn project as a dependency through submodules.
