AI Video Upscaling for Windows, Minus the Docker Headaches

What it does

enhancr is an Electron-based GUI for Windows that chains together video frame interpolation and AI upscaling models—think RIFE, RealESRGAN, and SwinIR—without forcing you to extract frames or build command-line pipelines. It reads frames directly from source video, encodes on the fly, and can batch-process a queue of files while showing a live preview.

The interesting bit

The app acts as a curated front-end for several inference backends: NVIDIA’s TensorRT for speed on GeForce cards, Tencent’s NCNN for cross-platform Vulkan inference on AMD and Intel GPUs, and Microsoft’s DirectML. It also bundles scene-detection and a realtime player with subtitle support, which suggests the author actually watches the output instead of just scripting it.

Key highlights

• Ships pre-packaged; the README emphasizes no Docker or WSL required to run inference.
• Supports chaining interpolation, upscaling, and restoration passes in one workflow.
• Can load custom ESRGAN models in ONNX and PyTorch .pth format and converts them automatically.
• Includes built-in scene detection to skip interpolation across hard cuts and reduce artifacts.
• Benchmarks in the README compare TensorRT and NCNN across mid-range to high-end NVIDIA cards; CAIN sees roughly a 4× speedup on an RTX 4090 when using TensorRT versus NCNN.

Caveats

• Windows is the only fully supported OS today; macOS and Linux builds are promised for the 1.0 release, and Apple Silicon support is still pending.
• Several listed engines—CAIN (NCNN), SwinIR (TensorRT), and SCUNet (TensorRT)—are marked as unreleased in the README.
• The “Pro” version and its embeddable Python environment are not distributed through the repository; only a free Windows installer is linked.

Verdict

Worth a look if you process video on Windows and want a GUI over VapourSynth pipelines. Skip it if you need a headless, cross-platform tool today or prefer to tune inference engines manually.