Upscayl

Upscayl

WHAT IS IT?

Upscayl is an open source desktop app that enlarges and enhances low-resolution images with AI models. Processing runs locally on the graphics card, on Linux, macOS 12+ and Windows 10+. The engine is Real-ESRGAN, run by upscayl-ncnn, an NCNN/Vulkan backend also released under AGPL-3.0.

WHY IS IT INTERESTING?

  • Seven built-in models: Upscayl Standard, Upscayl Lite, High Fidelity, Remacri, Ultramix Balanced, Ultrasharp and Digital Art, all 4x. You pick one based on the image type, photo or artwork.
  • Custom models: the app loads an external models folder. The upscayl/custom-models repo has more, and a guide explains how to convert your own PyTorch models to NCNN.
  • Batch Upscayl: select a folder and every image in it is processed with the same settings.
  • Double Upscayl: the image goes through the model twice, up to 16x. The UI says it plainly: above 4x (except 16x Double Upscayl), the app only resizes the image, without AI.
  • Output settings: PNG, JPG or WEBP, compression level, fixed output width instead of a scale factor, EXIF metadata copy, tile size and GPU selection.
  • Quick to pick up: drag and drop, paste from the clipboard (Ctrl+V / ⌘+V) and a four-step UI: image, model, output folder, run.
  • Packaged for every OS: Flathub, AppImage, AUR, Snap, DEB and RPM on Linux, Homebrew and the Mac App Store on macOS, an .exe installer on Windows.
  • CLI available: the upscayl-ncnn backend runs on its own from the command line.

Two limits to know. You need a Vulkan-compatible GPU: most integrated GPUs don't work, and the docs describe a CPU-mode workaround for Windows and Linux. Upscayl also does not fix blur or focus: it works on pixelated or low-resolution images, not on out-of-focus shots.

USE CASES

  • Enlarge an old low-resolution scan before printing it.
  • Upscale illustrations or artworks with the Digital Art model.
  • Process a whole folder of product photos or website visuals in one Batch Upscayl run.
  • Keep EXIF data on an upscaled photo collection, with JPG output for better metadata compatibility.
  • Plug upscaling into a script or pipeline through the upscayl-ncnn CLI.