Upscayl Desktop: free, open-source AI image upscaler that runs offline (GPU needed)
Upscayl says its Desktop app is completely free and open source, and does all processing on your own computer with no internet. It needs a Vulkan-compatible graphics card.

Official publisher artwork hosted by its publisher. Official source. Independent guide, not affiliated or sponsored. Artwork and trademarks belong to their owners.
What it is
Upscayl Desktop is an AI image upscaler. Its GitHub page says it lets you enlarge and enhance low-resolution images with AI, using Real-ESRGAN models. It runs on Windows, macOS and Linux. This post covers only the Desktop app, not Upscayl Cloud.
Is it free?
Yes, the Desktop app. Upscayl's download page says "Upscayl Desktop is completely free and open source. Download once, forever yours." The backend code is under the AGPLv3 licence, per its GitHub page. Upscayl Cloud is a separate online service with its own pricing page and paid plans. We did not review the Cloud plans, so nothing here applies to Cloud.
What your computer needs
- Upscayl's GitHub page lists Windows 10 and later, macOS 12 and later, and Linux.
- It needs a Vulkan-compatible graphics card (GPU). The page says many integrated graphics chips do not work, and that it "won't work with most iGPUs or CPUs". The README links a community workaround for Windows and Linux. That is not a promise of CPU compatibility or official support.
What it can and cannot do
- It guesses extra detail to make small or pixelated pictures larger. Upscayl's own FAQ says it cannot fix a blurred or out-of-focus image.
- Its site advertises up to 16x enlargement, batch upscaling, Double Upscayl and a choice of models. The Desktop download page also says there is no image-count limit. These are publisher claims, not an AiCoN benchmark. Hardware, model and output resolution still affect what your computer can process.
- Because the AI guesses detail, AiCoN advice: do not use an upscaled image as evidence or where exact detail matters, such as documents or IDs.
Getting started
- Check your operating system and Vulkan GPU support before downloading.
- Use Upscayl's official download page or the GitHub releases linked by its README. Choose the installer for your operating system.
- On macOS, the README describes opening the DMG and dragging Upscayl to Applications. On Windows, run the official EXE and follow its installation steps. On Linux, choose the package for your distribution; the README also describes a portable AppImage.
- AiCoN advice: start with one non-sensitive image and compare the result with the original before trying a batch. Keep the original.
The official wiki explains that default models natively output 4x; some other scale settings are made by reducing that 4x result. If a batch seems unfinished, the README says to wait for post-processing to finish. We did not verify installer screens or run an upscale.
Privacy
Upscayl's download page says Desktop does all the processing on your own computer and needs no internet connection. The privacy policy on its site is written for Upscayl Cloud, where the policy says it aims to retain input uploads for up to one day, while generated outputs are kept for 180 days or until deleted. It also warns that deletion timing can vary because of backups, caches and safeguards. Those Cloud terms do not describe Desktop. We have not audited the app or its network traffic.
India
Upscayl's pages do not name any country restriction for the Desktop download and no account is mentioned. Upscayl does not publish a list of countries, so we cannot confirm India availability from an official page. Check that the download works for you.
Frequently asked questions
Does it need internet?
Upscayl says Desktop processing is local and needs no internet connection.
Do I need a graphics card?
Yes. Upscayl says a Vulkan-compatible GPU is required and most integrated graphics will not work.
Is Upscayl Cloud free too?
Not covered here. Cloud is a separate service with paid plans on its own pricing page.
What changed since this post?
Download page, GitHub README/wiki, main website and Cloud privacy policy rechecked 12 October 2026. Confirmed Desktop free/local/offline claims and GPU limits. Added official installation routes, model-scale caveat and no-count-limit claim. Corrected Cloud retention wording to distinguish input uploads from outputs and avoid a deletion guarantee. No installation, performance, network or India availability test. Cloud prices and plans were not reviewed.
Sources: Upscayl download page, Upscayl on GitHub, Upscayl Cloud privacy policy, Upscayl feature page, Official wiki guide. Checked 12 October 2026.