AI upscaling for Windows that doesn't lock out Intel and AMD GPUs

What it does QualityScaler is a Windows desktop app that upscales and denoises images and videos using pre-trained models like BSRGAN and Real-ESRGAN. Written entirely in Python from backend to frontend, it orchestrates FFmpeg, ONNX Runtime with DirectML, and a CustomTkinter GUI into a PyInstaller-packaged binary. The author also sells it on Steam and itch.io, which is a neat trick for an open-source Python project.

The interesting bit Most AI upscalers treat CUDA as the one true faith; QualityScaler uses onnxruntime-directml to run on any DirectX 12-compatible GPU, including AMD and Intel cards, provided you have 4 GB of VRAM. It automatically tiles oversized images to dodge memory limits and supports pausing and resuming long video jobs—details that suggest actual dogfooding on real hardware.

Key highlights

• Fully offline: no internet connection or cloud handoffs required
• Broad format support, from HEIC and WebP to MKV, FLV, and MOV
• Multi-GPU support, including mixed integrated and dedicated setups
• Interpolation slider to blend the original and upscaled output
• Stop-and-resume for video encodes that take longer than your patience

Caveats

• Windows 10/11 only; Linux and macOS builds are not mentioned
• Requires a DirectX 12 GPU with at least 4 GB VRAM, plus 8 GB system RAM
• TTA upscale mode is still on the roadmap and not yet implemented

Verdict Windows photographers and video archivists with mid-tier AMD or Intel GPUs should give it a spin. If you are a Linux die-hard or need a headless batch-processing CLI, look elsewhere.