
Upscayl
Published by Upscayl
Free and Open Source AI Image Upscaler
Highlights
- Runs upscaling models locally, without sending images to a server
- Several bundled models trained for photographs, drawings and artefact-heavy sources
- Batch mode applying one model to every image in a folder
- Before-and-after slider for comparing the result against the original
- Needs a GPU with a working Vulkan driver to run at a practical speed
About this application
Upscayl enlarges images using neural network models that run on the machine itself. A picture is loaded, a model and an output folder are chosen, and the result is written out as a larger file. Nothing is uploaded — the processing happens locally, which also means the speed depends entirely on the hardware in front of you.
Several models are bundled, trained for different material: general photographs, drawings and digital artwork, and versions that also reduce compression artefacts. The interface shows a before-and-after slider over the result, and a batch mode runs one model across every image in a folder. Output format and scale factor are set before processing starts.
The models execute through Vulkan, so a graphics card with a working Vulkan driver is what makes this practical; on integrated graphics or without acceleration a single image can take a long time. The program is an Electron desktop application wrapping the upscaling binaries, published for Linux, Windows and macOS.
Where to get it
Linux
- Open on FlathubOpens in a new tabPackage repositoryFlathub
From the publisher
Install from a terminal
flatpak install flathub org.upscayl.Upscayl
We do not host this file, we do not scan it and we do not inspect its contents. Read what the source says, then decide for yourself.