Upscayl

The Upscayl desktop app is free and open source. The GitHub project carries the AGPL-3.0 license, and the home page describes the desktop download as free to keep. The pages reviewed describe no credit meter for the desktop app, so the only cost is your own computer and the time a large image takes to process.

Payment appears only with Upscayl Cloud, the separate web service. Its pricing page lists a Free Trial with 10 credits, a Pro plan at $24.99 per month with 300 credits, and Business and Custom plans sold by quote. The page does not say how many credits a single image costs, so the number of images a plan covers cannot be calculated from the page alone. That matters when you compare a Cloud subscription with a free download: the free route has a known cost of zero, while the paid route has a price per month but no published price per image.

Upscayl needs a Vulkan-compatible graphics card. The README states that most integrated GPUs do not work, while still allowing you to try, and the documentation repeats that many CPUs and integrated GPUs fail. Supported systems are Windows 10 or later, macOS 12 or later and Linux, with installers in several formats such as EXE, DMG, Flatpak, AppImage, Snap, DEB and RPM.

If an upscale fails on a laptop that has two graphics chips, the documentation suggests adjusting the GPU ID setting so the app targets the discrete GPU, then reinstalling and restarting. Check this before paying for Cloud credits only to avoid a hardware problem, since a paid plan will not repair a settings mistake on your own machine. Write down your graphics card model and operating system before you start, so a failed run can be compared with the compatibility list in the documentation.

Upscayl enlarges pixelated or low-resolution images using Real-ESRGAN models running on Vulkan. The README is direct about the limit: it cannot deblur an unfocused picture or correct focus. A sharp but small photo is a good input; a soft, out-of-focus one will only become a larger soft one.

Several models ship with the app, and the README names community models such as HFA2k, Remacri, UltraSharp and Ultramix Balanced, with more available from a custom-models repository. The home page also mentions upscaling up to 16x, a double-upscale pass, batch processing and face enhancement. Which model suits line art, photos or faces is not ranked by the provider, so compare on your own file.

• Sharp, small, pixelated source: the intended use.
• Blurry or out-of-focus source: not corrected, per the README.
• Many files at once: batch processing is listed for the desktop app.

The privacy policy covers Upscayl Cloud only and says uploaded images are used just to provide the service and are not used to train AI models. Input files are kept for up to 1 day after upload, and generated outputs for 180 days or until you delete them; the policy adds that backups and caches can delay deletion.

The terms say you keep full ownership of what you upload. Prohibited uses include infringing or harmful content and unauthorized scraping. The terms are governed by the laws of India, with arbitration in New Delhi, and name the operator only as Upscayl, so check them before sending client material. The desktop app processes files on your own machine, but the pages reviewed do not describe its data collection.

Pro adds rollover of unused credits, priority support, 6 months of storage, upscaling up to 256 megapixels and AI image generation. Business raises the megapixel ceiling to 512 and is quoted individually. Editorial assessment: someone upscaling a few images a month, on a machine with a discrete GPU, has no documented reason to pay, because the desktop app does the same core job for free.

Cancellation can be done at any time, but unused subscription credits are forfeited, and the terms exclude refunds for unused credits after you cancel or renew. Account history is deleted one month after cancellation. Download finished files before you stop paying.

1. Upscale it with the default model and again with one community model.
2. Open each result at 100% and compare edges, text and skin or texture detail with the original.
3. Repeat once with a deliberately out-of-focus copy of the same photo.
4. Note the processing time and the output file size.

Accept the result if edges stay clean and no invented detail appears in the text. Reject outputs with smeared texture or warped letters. The out-of-focus copy should stay soft, as the README says it will. Leave the observed result blank until the exercise is run.

For a design suite with built-in image tools, see Canva. For general image generation, see Adobe Firefly. For enlarging video rather than stills, see Topaz Video. Check your graphics card first.