LocalSR is a free desktop app for upscaling photos, cleaning up noise and blur, and enhancing video. Choose an image, pick Quick or Best, and compare the result with your original. The models run on your computer, and your photos stay there too.

Download · Try it in your browser · Website · User guide · Changelog

Original detail	LocalSR, 4× with Quick

A real result from SPAN NomosUni, with no added sharpening. Photo by Tim Ziegelbaum. Image credits and processing details.

LocalSR started out of necessity. For a long time I upscaled pictures with the HAT-L models, but only through the terminal. At some point I wanted a proper app for it, and most of what I found was subscription-based, really complicated or outdated. So the decision was clear. With the help of Claude and a clear idea of what I wanted, I built a super-resolution app that puts research-level models in one place and runs them on your own computer. The results are impressive, but they are not my work. My work is making these niche but incredible models a lot more accessible. They are meant to be used anyway: every model the app downloads is published under an open licence (Apache-2.0, BSD-3-Clause, MIT or CC BY 4.0 at the moment), and the app shows you each licence before it downloads anything.

• Enlarge photos and artwork by 2×, 3× or 4×. Quick is a good start, and Best spends more time on photo detail.
• Clean up noise, blur and JPEG artifacts first, then enlarge.
• Compare the original and the result with a slider, or hold M to see the original.
• Enhance video frame by frame, with the source timing, rotation and audio kept.
• Process a whole folder or a queue, and save a setup as a recipe.
• Edit with a prompt, with FLUX.2 klein or Qwen Image Edit. This is experimental, and so far only FLUX.2 klein has been tested, on a 16 GB Mac. About editing.

LocalSR also has a built-in benchmark. It times a real upscaling model on fixed test images, through the same inference engine the app uses for your own files, and reports how many megapixels per second your CPU or GPU can upscale. Local AI benchmarks can be hard to grasp, so it's good to have one built around a real use case.

It opens JPEG, PNG, WebP, TIFF, AVIF, GIF, JPEG 2000 and camera RAW files, and on the Mac also HEIC and JPEG XL. It saves PNG, JPEG, TIFF or WebP, or on the Mac also HEIC. Videos can be MP4, MOV, MKV, WebM and other common formats, and are saved as AV1, VP9 or FFV1, or on the Mac also as HEVC or ProRes. Video support.

Quick uses SPAN NomosUni and Best uses RealPLKSR NomosWebPhoto, both by Philip Hofmann. The library also has HAT, Real-ESRGAN, SwinIR, NAFNet, SCUNet and FBCNN, and each model shows its size, licence and source before you download it. Model guide · Model licences.

The download page has the installers for each platform:

• macOS v0.4.0-beta: Apple Silicon (M1 or newer) and macOS 14 or later. Intel Macs are not supported. Signed and notarized, and it updates itself.
• Windows v0.4.0-beta: Windows 10 or 11 on x86-64, as CPU, DirectML and NVIDIA CUDA editions. DirectML also runs on AMD and Intel graphics, built-in ones included, and Intel NPUs work on Windows and Linux. The installers are unsigned, so SmartScreen warns first: choose More info, then Run anyway.
• Linux v0.4.0-beta: x86-64 AppImages for CPU, AMD ROCm and NVIDIA CUDA. The CPU AppImage runs Intel graphics through OpenVINO once Intel's compute driver (intel-opencl-icd) is installed.

On a Mac you can also install it with Homebrew:

brew install --cask herrei/tap/localsr

The Mac app takes about 1.2 GB, because it brings its own Python and PyTorch. Models download when you choose them: Quick is 4.5 MB, most others are under 500 MB, and the editing models are 5 to 26 GB. I test upscaling on a Mac with 16 GB of memory.

LocalSR is only published on GitHub Releases, on herrei.github.io/localsr and through the Homebrew cask, which all use the same files. Each file has a SHA-256 next to it. The Mac app is signed as Developer ID Application: Hermes Reisner (Z2TU844D84); spctl -a -vv /Applications/LocalSR.app shows it.

• Upscaling reconstructs detail. Faces, text and fine patterns can look sharp and still differ from the real scene, especially at 4×. Results are saved as new files, so your originals stay untouched.
• Video is upscaled one frame at a time, so fine detail can flicker.
• Large images and long videos can take hours. Try a small one first.
• I test the Mac build the most; DirectML and CUDA less.
• On Windows and Linux, phone videos (H.264, HEVC, AAC) need an FFmpeg you install yourself. On the Mac they open through the system.

The full list is in Known limitations, and the testing details are in platform support.

There is no account, and your images and videos are never uploaded. At launch the app reads its update feed on GitHub Pages (for the beta, herrei.github.io/localsr/updates/beta.json) and can update itself, on the Mac, on Windows and as a Linux AppImage. After a check it sends one anonymous count with the app version, platform and channel to my Mac mini at macmini-ci.tail34a4e0.ts.net, without an install ID or cookies. Both are on by default. Turn either off in the LocalSR menu on the Mac, the Help menu on Windows and Linux, or the Software Update dialog; Check for Updates… still works when you ask. The Microsoft Store edition does neither.

Updates come from GitHub Releases, and the app checks their signature before it replaces anything. Models download from their authors on GitHub or Hugging Face when you choose them. That is all, and once your models are downloaded, LocalSR works offline. Privacy details.

Why is it free? It's my own project, made alongside uni. There is no paid version, no account and no advertising, and the code is MIT licensed.

Can I use my own models? Yes, compatible .safetensors files work as a custom checkpoint. .pth, .pt and .ckpt files can run code when they load, so the app only uses one after you tick Allow this pickle-based checkpoint. Do that only for files from a publisher you trust.

How do I uninstall it? On a Mac, move LocalSR to the Trash and delete ~/Library/Application Support/LocalSR and ~/Library/Caches/LocalSR. On Windows, uninstall it under Apps and delete %LOCALAPPDATA%\LocalSR. On Linux, delete the AppImage and ${XDG_DATA_HOME:-~/.local/share}/LocalSR. The guide has the details.

You need Python 3.11, Node.js, a stable Rust toolchain and Tauri's platform prerequisites.

./local-ci.sh setup
cd desktop && npm run tauri -- dev

The interface is Svelte in Tauri, with a Rust host for the queue, settings and updates, and a separate Python worker that runs PyTorch and spandrel. Run ./local-ci.sh for the local checks. Desktop guide · Worker protocol · Command line · Releasing.

Please report bugs on GitHub Issues with your app version, model, device and the steps that led to the problem. Copy diagnostics in the app adds the hardware details. For security reports, see SECURITY.md. For anything else: [email protected].

LocalSR is MIT licensed. Models keep their authors' licences, shown in the app before download. Bundled libraries are listed in THIRD_PARTY_NOTICES.md, and each release includes the source of its LGPL components.