Upscayl: a free, open source AI upscaler that runs on your own computer

There is a photo in your camera roll that you keep meaning to print. The moment was perfect, but at full size the face is soft, the background is a smear of noise, and the pixels do everything they can to remind you that it came from a phone. You have probably met this problem under other names. A scanned family photo that came back from the lab looking gray and thin. A game screenshot that turns to mush the moment you post it full width. A piece of fan art you want to frame but can only print at the size of a credit card.

Most upscalers you will find online handle this the same way. Upload your image, wait for a cloud server to chew on it, download the result, and hope the subscription you just signed up for is worth it. Upscayl flips that. It is a desktop app that runs the upscaling on your own computer, and the whole thing is free and open source. You download it once from the official Upscayl site, point it at a folder of images, and let it work. Nothing leaves the machine unless you tell it to.

The project lives at the Upscayl GitHub repository, where anyone can read the code, file an issue, or watch the team ship updates. It is built in TypeScript as an Electron app, the same tech family as apps like Slack or VS Code. It runs on Windows, macOS, and Linux. It carries around 49,000 stars on GitHub, which for a desktop utility is a genuinely loud crowd, and the repository has been pushed to recently enough to call it alive.

What it actually does

AI upscaling works by teaching a model what a sharp version of a blurry thing usually looks like. The model was trained on pairs of small images and the crisp versions of those same images, so when it sees a low resolution photo it reconstructs plausible detail instead of just stretching the pixels. That is the whole trick, and it is why an upscaler can make a photo look genuinely sharper than a plain resize ever would. It cannot bring back a face that was out of focus to begin with, but it can turn a soft 800-pixel image into something that holds up at poster size.

Upscayl does this locally, and it bundles several different upscaling models you can switch between. Different models lean in different directions. Some are good at photographs and keep skin and fabric looking natural, while others are tuned for digital art, anime style illustration, and crisp line work. The interface lets you pick a model, choose your target scale, and run the job. The site advertises upscaling up to 16x for large jobs, though for most photos 2x or 4x is the sweet spot. If a single pass does not get you there, Upscayl has a double upscale option that runs the image through twice in a row, which helps when you need a big jump in resolution.

Two workflow features make this more than a single photo toy. Batch processing lets you drop a whole folder of images and let Upscayl work through them one after another, which is exactly what you want when you have a camera roll full of old scans. And because everything runs on your own hardware, a folder of family photos never has to be uploaded to a third party server to get fixed.

Where it earns its place

Restoring old scans is the classic use. A photo of a photo, or a scan of a family album, often comes back grainy and low resolution. Run it through Upscayl with a model aimed at photographs, and you usually get a clean version you can actually frame without squinting. For creators working in digital art, the anime and illustration models are a better fit. They keep line work tight and avoid that smudge you sometimes get when a photo model is forced onto painted artwork. If you build or edit game assets, product mockups, or social media banners, Upscayl is a fast way to take a concept at preview resolution and push it up to something that reads as intentional instead of compressed.

The privacy angle is real for a specific kind of user. A photographer working with a client, a designer handling someone else's brand assets, or a developer prototyping with screenshots that contain interface details. In those cases, the fact that the image never has to touch a cloud queue is a genuine advantage, and one that is hard to match with a subscription service.

What it costs, and what to know before you install

The desktop app is free and open source under the AGPL 3.0 license. You download a build for your operating system from the site or the releases page on GitHub, install it, and you own it. There is no credit meter watching your shoulder. The project has also launched a paid Upscayl Cloud tier, which adds cloud storage and some extra AI features like face enhancement and image generation. That cloud part is optional. The core desktop app is where most people will live, and it is the version that stays on your machine.

The tradeoff is that "local" also means "your hardware does the work." If you are on a modern laptop with a decent graphics card, large images will process in seconds to a few minutes. On an older machine, big 4x or 16x jobs can drag, and a folder of two hundred photos will take a chunk of your afternoon. If your graphics card is doing other things at the same time, expect it to compete for attention. None of that is a dealbreaker, but it is the honest part of the "runs on your computer" pitch.

Honest limits

Upscayl is a single purpose tool, and it does not pretend to be more. It will not color correct a badly white balanced photo, and it will not fix a composition that does not work. It fixes resolution and sharpness, and it does that well. If your photo is blurry because the camera moved, upscaling gives you a cleaner blurry image, which is better but not a rescue. The same goes for tiny text. Upscaled text is less jagged, but it is still the same text, so do not rely on it to make a 20-pixel caption legible.

You should skip it if you are doing this once a year and would rather not install anything. For a one off photo, any of the web based upscalers will get the job done in a couple of clicks. Upscayl starts to make sense the moment you have more than one image, a folder of them, or a reason to keep the originals private. That is also where the open source license starts to matter if you are building on top of it. AGPL is the strongest common open source license, so if you plan to fork it and ship a closed product, read the fine print first.

How it compares, and who should try it

The web based upscalers like Lets Enhance, Clipdrop, and Topaz are all legitimate. They are usually faster to start because there is nothing to install, and some of them offer extra one click fixes that go beyond pure upscaling. The cost is a credit or subscription model, and your images live on someone else's server for a little while. Upscayl's position in that landscape is simple. It is the free, private, batch friendly option for people who have a steady stream of images and do not mind a one time install. If that sounds like your situation, it is worth an afternoon.

Who should actually try it. If you are a creator with a backlog of old photos, scans, or artwork you want to save, install it and run a batch. If you are a designer or developer who handles other people's images and privacy matters, it is the safest default on your machine. If you only ever upscale a single image once a month, a web tool is the easier path and you will not miss what you are missing. The one place I would not reach for Upscayl is a professional print pipeline where color grading and fine retouching are the whole job. It gives you a cleaner canvas, but you still need the editor to do the rest. For everyone else with a folder of soft images, it is a genuinely good tool, and the free and open source part means it keeps running long after the trial period of a subscription would have run out.