Video2X

Video2X is a machine learning-based lossless video super-resolution and frame interpolation framework designed to enhance video quality by increasing resolution and frame rate without introducing noticeable artifacts. Unlike traditional interpolation methods that simply duplicate or blend frames, Video2X employs trained neural network models—such as waifu2x, SRMD, and RealSR—to intelligently upscale each frame, preserving fine details and textures. The frame interpolation component uses models like RIFE (Real-Time Intermediate Flow Estimation) to generate intermediate frames between existing ones, resulting in smoother motion and higher temporal resolution. The framework operates as a comprehensive processing pipeline: it first extracts individual frames from a source video (using FFmpeg), then applies the selected upscaling and/or interpolation models to each frame, and finally re-encodes the processed frames back into a video file. Users have control over the model selection, scaling factor (e.g., 2x, 3x, 4x), output codec, quality settings, and batch processing parameters. Video2X supports GPU acceleration via CUDA, DirectML, Vulkan, and OpenCL—allowing users to leverage NVIDIA, AMD, Intel, or Apple Silicon GPUs for faster processing. It also offers CPU fallback for systems without compatible hardware. Practical use cases include restoring old or low-resolution footage (e.g., VHS captures, early digital videos, game recordings), enhancing anime and cartoon content (where waifu2x models excel), preparing high-resolution material for large displays or projectors, and smoothing choppy video streams for archival or creative projects. The software is available on Windows (both command-line and Qt6 GUI versions) and Linux (command-line and GUI). The Qt6 GUI provides a grap