This project provides ML filter runtimes for VapourSynth video processing, supporting hardware-accelerated inference across CPUs, NVIDIA GPUs (CUDA/TensorRT), Intel GPUs (OpenVINO), AMD GPUs (MIGraphX/NCNN/Vulkan), and Apple Silicon (CoreML). It bundles optimized runtimes for popular models including waifu2x, RealESRGAN, Real-CUGAN, RIFE, SCUNet, DPIR, and ArtCNN. A unified Python wrapper is provided to simplify backend selection.