Free AI image upscaling that actually runs on your own GPU

What it does

Upscayl is an Electron-based desktop application that enlarges and enhances low-resolution images using Real-ESRGAN models through a Vulkan backend. It guesses missing details to upscale images while keeping everything on your local machine. The project also maintains a separate CLI backend called upscayl-ncnn for users who prefer terminals to GUIs.

The interesting bit

The project is essentially a polished delivery mechanism for open-source AI models—think of it as the friendly packaging around the hard math. It supports custom model conversion and batch processing, and the entire inference backend is fully open source under the AGPLv3 license.

Key highlights

• Performs all image inference locally using Real-ESRGAN models and a Vulkan backend
• Requires a Vulkan-compatible discrete GPU; most integrated graphics are explicitly unsupported
• Supports custom models and batch upscaling, with a separate repository for additional models
• Cross-platform desktop builds for Linux, macOS, and Windows
• Open-source backend (upscayl-ncnn) licensed under AGPLv3

Caveats

• It will not de-blur, sharpen focus, or fix out-of-focus images; it only enhances existing low-resolution or pixelated sources
• GPU compatibility is a hard requirement, and the project acknowledges that CPU or iGPU fallback is generally not available
• Interrupting a batch job can leave images unprocessed or at the wrong scale because post-processing happens only after all upscaling finishes

Verdict

Grab it if you have a Vulkan-compatible GPU and a backlog of pixelated images to enlarge. Look elsewhere if you rely on integrated graphics or need to rescue blurry, out-of-focus shots.