Upscayl

Free open-source AI image upscaler for desktop.

Image Generation

Overview

Upscayl is the free, open-source, desktop AI image upscaler that a lot of working professionals quietly use before reaching for anything paid. It runs entirely on your machine, processes images with no cloud round trip, has no subscription, no credit system, and no telemetry. It ships as a standalone app for Windows, Mac, and Linux. It is licensed AGPL. The whole thing is built by a small open-source team led by Nayam Amarshe and TGS963, funded by donations and a Pro version, and shipped through GitHub Releases like a normal open-source project. In an image-tool market that has moved almost entirely to SaaS subscriptions and per-image credits, Upscayl is the deliberate counter-argument.

The underlying technology is not proprietary. Upscayl wraps Real-ESRGAN and a set of related open-weight upscaling models — the same models that power a lot of the enhancement inside more expensive tools — behind a clean GUI. That framing is important. Upscayl is not competing with Magnific's creative reimagination or Krea's realtime canvas. It is a faithful upscaler: take an image, enlarge it up to 16x, sharpen and denoise while preserving the original content, output a bigger clean version. No hallucinated detail. No slider labeled Creativity. No prompts. Just resolution.

That narrow focus is Upscayl's superpower. Photographers who need to salvage a small JPEG, hobbyists restoring old family photos, artists blowing up sketches, developers scaling assets in a game pipeline, anyone with a stack of low-resolution images to make bigger — Upscayl handles all of them, offline, for free, in seconds per image on a modern GPU. The tool is downloaded well past three million times on GitHub as of 2026 and has become the reference free option in the space.

Where Upscayl stops mattering is anywhere you want the enhanced image to be more than a bigger clean version of the input. If you want to reinterpret detail, add photorealism to a sketch, or push a soft image toward editorial quality — Magnific, Krea, or Topaz Photo AI are the tools. Upscayl is faithful. Faithful is the entire point.

Key Features

  • Fully offline processing. No cloud, no upload, no account, no telemetry. Runs entirely on your local GPU or CPU. The specific feature that makes Upscayl usable on private, confidential, or client-NDA imagery.

  • Multiple upscaling models bundled. Digital Art, Ultrasharp, Standard, Remacri, Ultramix, and specialized models for anime, illustration, and photography. Switching models takes a click; each has a different character. Working users learn 2-3 favorites for their specific content.

  • Up to 16x upscaling. Practical outputs are typically 2x-4x; 8x and 16x work but eat processing time and often exceed what the source image contains as recoverable detail. Realistic use is 4x on most images.

  • Batch processing built in. Point at a folder, pick a model, pick a scale factor, walk away. Batch is the reason Upscayl becomes indispensable inside a professional workflow — you leave it running overnight on a stack of a hundred images.

  • Face Enhance option. Optional GFPGAN pass that specifically restores facial features. Useful on old family photos, screengrabs, and low-resolution portraits. Not perfect — occasional plastic-skin artifacts — but generally usable.

  • Custom model support. Drop in third-party Real-ESRGAN and NCNN models from the community and use them inside Upscayl. The extensibility that pro users care about.

  • Format flexibility. JPG, PNG, WebP input; JPG, PNG, WebP output. Configurable compression on output. Alpha channels preserved.

  • CLI in Pro. Command-line interface for scripting into pipelines. The Pro tier's most useful feature for developers who want to use Upscayl in an asset build step.

  • Cross-platform desktop. Native Windows, Mac (Intel and Apple Silicon), and Linux binaries. AppImage for Linux users. The rare AI tool that treats Linux as a first-class platform.

Pricing

Upscayl is genuinely free. The Pro tier exists but is optional.

Version Price Includes
Upscayl (free) $0 All models, batch, offline, unlimited use
Upscayl Pro $9.99 one-time CLI, priority updates, additional models, commercial license
Cloud (optional) ~$5/month Server-side upscaling for weak-GPU machines

The free version is not crippled. It includes every core model, batch processing, all scale factors, and unlimited use. There is no watermark, no daily limit, no credit system. This is important: unlike most tools that use free-tier limits as a funnel, Upscayl's free version is the actual product for the vast majority of users.

Pro at a one-time $9.99 unlocks CLI access, early updates, some additional model bundles, and grants explicit commercial-use licensing. For anyone using Upscayl commercially — even on a small scale — the Pro purchase is the honest way to support the project and clarify licensing. It is a one-time purchase, not a subscription.

Cloud is optional and priced at roughly $5/month for people whose local machine cannot run the models fast enough. The core project remains local-first — Cloud is a convenience, not the pricing model.

Pros and Cons

Pros

  • Genuinely free, genuinely offline. No subscription, no credits, no telemetry.
  • Batch processing is unlimited and local. Overnight runs on stacks of hundreds of images.
  • Multiple model options mean you actually match the tool to the content — anime, photo, illustration, art all have their preferred models.
  • Cross-platform including Linux as a first-class target.
  • Runs on modest hardware. A four-year-old GPU handles standard upscaling in seconds per image.
  • Open-source under AGPL. The code and the models are yours. The project cannot silently change its pricing on you.

Cons

  • Faithful upscaler only — no creative reimagination, no prompt control. If you want Magnific-style creative detail, this is not the tool.
  • Ceiling on quality is capped by the underlying models. Real-ESRGAN and its relatives are good, but 2026 has moved past them at the top of the market.
  • Face Enhance is hit-or-miss. GFPGAN's plastic-skin artifacts are still occasional even on the tuned Upscayl builds.
  • No cloud sync, no library, no versioning. It processes files on disk and writes files on disk. That is deliberate but it is not a workflow tool.
  • No API in the free version. Pro CLI covers most pipeline use, but developer integration is not as clean as a real REST API.
  • Local GPU requirement. On CPU-only machines or old integrated graphics, the tool is slow enough to hurt.

Best Use Cases

  • Photographers salvaging small JPEGs. Old photos, phone screenshots, client-supplied low-res files. Upscayl at 4x on the right model produces genuinely useful outputs.
  • Family archivists restoring old photos. With Face Enhance carefully applied, old prints and small scans become viewable at modern sizes.
  • Game developers and asset teams scaling texture and sprite assets in a build pipeline. Pro CLI plus a folder watch script covers this.
  • Digital artists blowing up sketches and reference images for further work in Procreate or Photoshop.
  • Anyone with confidentiality or NDA constraints on their imagery. Local-only processing removes the entire class of cloud-privacy questions.
  • Anyone on a budget who wants the 80% solution for zero dollars. Upscayl is honestly that.

Alternatives

  • Magnific AI. The paid ceiling. Creative reimagination and dramatic detail invention, at $39-99/month. See /tools/magnific-ai. The right choice when the goal is editorial or portfolio-grade output.

  • Krea AI Enhancer. The middle ground. Cheaper than Magnific, more capable than Upscayl, cloud-based. See /tools/krea-ai.

  • Topaz Photo AI. The paid photographer's tool. Denoise, sharpen, and upscale with more control than Upscayl and a serious model set. Around $199 one-time. The right choice when you want a professional native app and can spend the money.

  • Real-ESRGAN direct. If you are comfortable with Python and a command line, the models Upscayl wraps are usable directly. Zero GUI, more flexibility, more setup cost.

  • Adobe Firefly Super Resolution / Photoshop. Built-in inside Camera Raw. If you already pay for Creative Cloud, it covers most casual upscaling with reasonable quality.

For a broader look at where generative image tools sit relative to each other, see Midjourney vs DALL-E 3 vs Stable Diffusion and /tools/midjourney.

Getting Started

  1. Download from upscayl.org or GitHub Releases. Pick the installer for your OS. Windows and Mac installers are signed. Linux uses AppImage or the .deb / .rpm packages depending on distro.
  2. Try three models on the same image. Real-ESRGAN, Ultrasharp, and Digital Art on a single test file. See which model handles your content best before batching a folder — the wrong model choice on 500 images wastes hours.
  3. Batch a real project. Point Upscayl at a folder of 20-50 images and run it overnight. This is where the tool proves its value versus paying per-image on a cloud service.

FAQ

Is Upscayl really free? Yes, in a way that is unusual for the AI space. Free of subscription, free of credits, free of watermark, free of telemetry. The Pro tier is a one-time $9.99 and unlocks CLI plus commercial-use license.

Can I use Upscayl output commercially? For the free version, the AGPL license and the underlying model licenses generally permit commercial use, but the Pro purchase makes commercial licensing explicit and supports the project. If you use it commercially, buy Pro.

How does Upscayl compare to Magnific? Different tools. Upscayl is a faithful upscaler; Magnific is a creative reimaginer. For 4x sharpening of a clean source, Upscayl is 90% of the quality at 0% of the cost. For creative reinterpretation, Upscayl is not the tool at all.

Does Upscayl work without a GPU? Yes, on CPU, but slowly. A 2018-era GPU is enough to make it fast enough for real use. On CPU-only machines, batch runs of dozens of images may take hours.

Which model should I use? Real-ESRGAN Standard is the safest general default. Digital Art for illustration, sketches, and stylized art. Ultrasharp for photography where you want more aggressive detail recovery. Face Enhance only when you specifically need faces improved.

Does Upscayl phone home? No. The desktop app has no telemetry. All processing is local. The only network calls are optional update checks and the optional Cloud feature.

Verdict

Upscayl is the correct answer to "what should I use to upscale an image" when the honest answer is that the image just needs to be bigger and sharper. It is free, offline, unlimited, cross-platform, and open-source, and it produces output quality that covers the vast majority of practical upscaling needs. Photographers, hobbyists, game developers, digital artists, and anyone with a stack of low-resolution files should download it before paying anything. The Pro tier at $9.99 one-time is a fair way to support the project if you use it commercially. Upscayl is not the tool for creative reimagination, editorial print output, or dramatic detail hallucination — Magnific and Krea's Enhancer own that ground. But for the everyday "make this bigger without ruining it" use case that used to require Topaz Gigapixel and a $199 purchase, Upscayl is now the default. The best free tool in AI image processing in 2026 by a comfortable margin.