Upscayl

Upscayl is an open-source AI image upscaling tool for desktop, started by Nayam Amarshe and TGS963 in July 2022. With 46k+ stars on GitHub, it brands itself as "the #1 free and open source AI image upscaler" — a phrase used across the README and the official site. It is not part of the SD-WebUI family, but it sits in the same open-source desktop imaging category.

Core capabilities focus on a single point: AI upscaling. First, local inference on Real-ESRGAN: the backend uses upscayl-ncnn (also AGPL-3.0 licensed) and runs multiple models on the GPU via Vulkan — Real-ESRGAN, UltraSharp, Remacri, Ultramix Balanced, High Fidelity, and others. Second, batch processing: Batch Upscayl mode upscales a folder of images and applies post-processing when finished. Third, 2x and 4x upscaling options with configurable output format and quality. Fourth, multi-platform native support: Linux ships Flatpak, AppImage, .deb, .rpm, and AUR packages; macOS ships dmg and Homebrew Cask; Windows ships an exe installer; iOS is on the App Store.

Use cases include upscaling old low-resolution photos or illustrations, post-upscale of AI-generated images, bumping design files up a tier before delivery, and privacy-sensitive workflows where images must stay on-device. The onboarding bar is low: 1-2-3 steps from download to first run on desktop. The hardware baseline is a Vulkan-capable discrete GPU; the README states "Most iGPUs and CPUs will not work" but encourages trying.

Licensed under AGPL-3.0; the project accepts Buy Me a Coffee donations. The command-line tool lives in a separate repository, upscayl-ncnn. If you can only remember one open-source AI image upscaler, Upscayl is the de facto desktop standard.