Upscayl is a free, open-source AI image upscaler that runs locally on Windows, Mac, and Linux.
Upscayl wraps the Real-ESRGAN family of models in a polished desktop app, letting anyone upscale images without uploading them to the cloud. Being open source (GPL), it is free for personal and commercial use, and it works offline — a decisive advantage for private or sensitive material. The app supports batch upscaling, multiple AI models, and double upscaling for extreme enlargement, with active community development and model packs.
Completely free. Upscayl is community-funded open-source software; optional donations support development. No credits, no watermarks, no subscriptions.
Privacy-conscious users, Linux enthusiasts, and anyone who wants unlimited upscaling without paying or uploading images.
Privacy-conscious users upscale sensitive documents locally. Linux enthusiasts enhance screenshots and wallpapers. Hobbyists improve game textures and anime art without paying per image.
Try the ultrasharp and ultramix models on photos, and switch to anime-optimized models for illustration. Batch-process folders overnight for large archives.
Upscayl fits into any offline workflow: drop a folder in, choose a model, and process locally overnight. Custom model support means enthusiasts can swap in specialized upscalers for anime, photos, or text, all without sending anything to external servers or paying per image.
Yes, entirely free and open source under the AGPL license.
There is no hosted API since processing is local, but the underlying models can be self-hosted.
It matches paid tools on typical photos for free; cloud tools add generative detail Upscayl does not invent.
No, after installation everything runs locally — images never leave your machine.
Yes, custom Real-ESRGAN style models can be loaded into the app.
Yes, the software is free for personal and commercial use under its license.
It is community-supported open-source software, with optional donations funding development.
Common raster formats including JPG, PNG, and WebP are supported as input, with enhanced results exportable at standard formats for further use.