Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Upscayl

Free desktop app that enlarges images offline with Real-ESRGAN-style models on any Vulkan GPU.

Reviewed by Sheela · Updated

View on GitHub Open source

Our rating

3.8/5

Upscayl remains the simplest free way to enlarge a photo without uploading it, but the app you download today is the one from Christmas 2024.

What is Upscayl?

Upscayl's last release, v2.15.0, shipped on 25 December 2024, so no release has come out in more than 22 months. The main branch has since gained about 50 commits, mostly new interface translations (Polish, Hungarian, Italian, Ukrainian, Arabic and others), a Copy Metadata option and small fixes; its last commit was on 27 March 2026. The push on 5 October 2026 went to an unreleased upgrade-codebase branch, and a v3 pull request (#1397) was closed without merging in September 2026.

The app itself is an Electron and TypeScript front end for upscayl-ncnn, a Real-ESRGAN engine that runs on the NCNN framework through Vulkan. You drop in an image or a folder, choose one of seven bundled 4x models (Standard, Lite, Digital Art, High Fidelity, Remacri, Ultramix Balanced and UltraSharp) or your own NCNN model, and save the result. Everything runs on your own GPU; nothing is uploaded.

Builds cover Linux (AppImage, deb, rpm, Flathub, Snap, AUR), macOS 12 or later (DMG, Homebrew, Mac App Store) and Windows 10 or later. The code is AGPL-3.0. The same team sells Upscayl Cloud, a separate paid web service with larger outputs, face enhancement and image generation.

Is Upscayl worth it?

Upscayl does one job, enlarging images on your own graphics card, and does it with less friction than anything else that is free. The repository has about 51,500 GitHub stars, and the Windows installer for v2.15.0 alone shows about 11.4 million downloads on the release page. The trade-off is momentum: v2.15.0 from 25 December 2024 is still the newest release, the Mac App Store build was last updated on 27 December 2024, and Flathub still serves 2.15.0 to about 8,193 downloads a month. Work since then has gone into translations on main and an unreleased redesign on side branches, while pull requests for recursive batch processing (#1430) and batch progress (#1435) were closed unmerged in July 2026.

The models are the honest limit on quality. All seven bundled models are 4x Real-ESRGAN-family networks, including community favourites Remacri, UltraSharp and HFA2k (shipped as High Fidelity). The site advertises up to 16x, but according to the wiki the default models only upscale natively by 4x and other scales are emulated by shrinking the 4x result; Double Upscayl simply runs the model twice. Custom NCNN models (.bin plus .param) load from a folder, and PyTorch models can be converted with the documented guide. The FAQ is candid that Upscayl sharpens low-resolution or pixelated images but cannot fix blur or missed focus, which is where generative upscalers pull ahead.

Hardware is the other gate. Upscayl needs a Vulkan-compatible GPU, and the README warns that many integrated GPUs and all CPUs are unsupported. The community compatibility list says any Vulkan 1.3 GPU should be fine and reports Intel HD 620 and 630, Iris and AMD Vega 8 and Vega 10 working. The only CPU route is an unofficial Windows trick from issue #390 that swaps in the SwiftShader software renderer, which its author called slow; on Linux it is unconfirmed and on macOS there is none.

Against paid rivals the case is mostly price and privacy. Topaz Gigapixel costs $149 a year upfront for the Personal plan or $499 for Pro, adds generative models, Photoshop and Lightroom plugins and optional cloud rendering, and limits Personal to organisations under $1M in revenue. Bigjpg's free online tier stops at 4x, 5MB files and 20 images a month, and every image goes to its servers. Upscayl Cloud is the maintainers' own answer for bigger jobs, with outputs up to 256MP on Pro, but it does not publish how many credits an image costs and lists no video upscaling.

The AGPL-3.0 licence covers the app and the upscayl-ncnn engine, not the images you make, so freelancers and companies can use the output commercially. It bites only if you ship a modified Upscayl or run it as a service for others, in which case you must publish your changes; the README credits the community model authors but does not state those models' own licences.

Where it earns its place

  • A small online shop owner with 300 supplier product photos at 800 pixels wide

    Enlarge the photos enough for zoomable product pages without paying per image

    Point Batch Upscayl at the folder, choose the Standard or UltraSharp model and let it run on the office PC's GPU. It costs nothing per image; the alternatives are Bigjpg's paid plans at $6 to $22 for a few months of credits, or Topaz Gigapixel from $149 a year.

  • A family-history volunteer scanning 1990s prints for a local archive

    Make low-resolution scans printable while keeping private photos off cloud services

    Run the Flathub or DMG build offline and save upscaled copies alongside the originals. Pixelated scans improve; blurred or badly focused prints will not, per the FAQ, so they still need retouching or a generative tool.

  • A developer building an internal image pipeline on a Linux server with an AMD GPU

    Upscale thumbnails in bulk from a script rather than a desktop app

    Call upscayl-ncnn, the AGPL-3.0 command-line engine behind the app, with an NCNN model and a GPU ID. Internal use is fine; offering it to outside users as a modified hosted service triggers the AGPL's source-publication duty.

Before you commit

Setup
Download the AppImage, DMG or EXE from GitHub or upscayl.org, or install from Flathub, Snap, AUR, Homebrew or the Mac App Store ($12.99, 571.5 MB). Windows builds trigger a SmartScreen warning, and the Flathub build is x86_64 only. Upscaling starts as soon as the app opens if the GPU is supported.
Integrations
None beyond the file system and clipboard paste. Custom NCNN models load from a folder in Settings, a GPU ID setting picks the card (or several, which the wiki says do not share load evenly), and upscayl-ncnn works as a separate CLI.
Data and privacy
Images are processed locally. v2.15.0 added an option to share usage stats such as GPU model and daily users, which the release notes say include no personal data, and auto-update checks can be switched off. The Mac App Store label reads Data Not Collected. Upscayl Cloud is separate and does upload images.
Cost at scale
Free at any volume on your own hardware; the cost is a Vulkan-capable GPU. Upscayl Cloud's Pro plan is $24.99 for 300 credits a month with outputs up to 256MP; Business allows 512MP; tiers run from 100 to 5,000 credits, annual billing is advertised at up to 40% off, and credits per image are not published.
Limits
47 open issues and 1 open pull request. No CPU mode beyond the unofficial SwiftShader swap, native 4x models only, and no video upscaling in the desktop app. Cancelling Upscayl Cloud forfeits unused credits, and history is deleted after a month.

What’s good

  • Free, offline and no account; images never leave the machine
  • Seven bundled 4x models plus custom NCNN models
  • Batch upscaling of whole folders
  • Runs on NVIDIA, AMD and Intel GPUs through Vulkan, on all three desktop systems

What’s not

  • No release since v2.15.0 in December 2024
  • Needs a Vulkan GPU; no official CPU mode and many integrated GPUs fail
  • Native 4x only; other scales are resized from the 4x result
  • Cannot recover blurred or out-of-focus photos

Best for

Photographers, designers and archivists with a discrete or recent integrated GPU who want free, private 4x enlargements of sharp but small images.

Upscayl alternatives

Upscayl: common questions

Is Upscayl worth it?
Upscayl remains the simplest free way to enlarge a photo without uploading it, but the app you download today is the one from Christmas 2024. We rate it 3.8 out of 5.
How much does Upscayl cost?
Free and open source (AGPL-3.0) from GitHub and the website; $12.99 on the Mac App Store. Upscayl Cloud offers a 10-credit free trial and Pro at $24.99 for 300 credits a month.
What are the downsides of Upscayl?
No release since v2.15.0 in December 2024; needs a Vulkan GPU; no official CPU mode and many integrated GPUs fail; native 4x only; other scales are resized from the 4x result; cannot recover blurred or out-of-focus photos.
Who is Upscayl best for?
Photographers, designers and archivists with a discrete or recent integrated GPU who want free, private 4x enlargements of sharp but small images.
How much does Upscayl cost for a team or in production?
Free at any volume on your own hardware; the cost is a Vulkan-capable GPU. Upscayl Cloud's Pro plan is $24.99 for 300 credits a month with outputs up to 256MP; Business allows 512MP; tiers run from 100 to 5,000 credits, annual billing is advertised at up to 40% off, and credits per image are not published.
Is Upscayl safe for company data?
Images are processed locally. v2.15.0 added an option to share usage stats such as GPU model and daily users, which the release notes say include no personal data, and auto-update checks can be switched off. The Mac App Store label reads Data Not Collected. Upscayl Cloud is separate and does upload images.
What are the best Upscayl alternatives?
The closest alternatives we have reviewed are ComfyUI and InvokeAI.

Answers checked Oct 31, 2026.