Upscayl

Free and open-source AI image upscaler for desktop

Freemium· Desktop: Free and open source. Cloud: Paid subscription (tier pricing on site).Image GenerationReal-ESRGAN and related ncnn-Vulkan super-resolution models (plus custom user models)

Visit website →

Best for

Photographers, designers and privacy-conscious users who want a free, offline, cross-platform way to upscale photos and artwork on their own GPU without subscriptions or uploads.

Skip if

Users who need face-restoration, generative fill, or the absolute state-of-the-art fidelity that paid tools like Topaz Gigapixel or Magnific deliver on portraits and text.

Upscayl is a free, open-source desktop application that uses AI super-resolution models (the Real-ESRGAN family and related ncnn-Vulkan ports) to enlarge and clean up low-resolution or blurry images. It bundles several purpose-built model weights so users can pick between a general photo model, a digital-art / illustration model, a sharpening pass, and specialty variants tuned for skin, text, or heavily compressed sources. The application runs entirely on the local machine using Vulkan for GPU acceleration, which means it works cross-platform on Windows, macOS (Intel and Apple Silicon) and Linux without needing a CUDA-capable NVIDIA card.

Core workflows include single-image upscaling at 2x, 3x, 4x or up to 16x (via a double-upscale pass), batch processing of a folder, and configurable output format and compression. There is a companion paid Upscayl Cloud service for users who lack a capable local GPU or want to run jobs from a browser or mobile device, plus a growing custom-models ecosystem where power users can drop in their own .bin/.param ncnn files.

The tool is aimed at photographers restoring old scans, print designers who need to blow up a small source asset, e-commerce operators cleaning product photography, retro-game artists upscaling sprites and textures, and anyone who wants a no-subscription, no-cloud alternative to Topaz Gigapixel or Adobe Super Resolution. Because it is local, offline and MIT/AGPL-licensed, it is also popular in privacy-sensitive workflows such as legal, medical archive digitisation and journalism.

Editor's take

Upscayl is the answer whenever someone asks for a free Topaz Gigapixel alternative. It won't beat commercial upscalers on tricky faces, but for batch-restoring old scans, blowing up product shots, or cleaning up digital art it's remarkably good — and the fact it runs locally on any GPU, MIT-licensed, with no account required, makes it the default recommendation.

— The AI Tool Bible editorial team

Pros

✅ Genuinely free and open source under a permissive license, with active GitHub development
✅ Runs fully offline on your own machine — no images ever leave your device in the desktop app
✅ Cross-platform (Windows, macOS Intel and Apple Silicon, Linux) via Vulkan, so it works on AMD and Intel GPUs too, not just NVIDIA
✅ Ships multiple task-specific models (photo, digital art, sharpen, high-fidelity) rather than a single one-size-fits-all upscaler
✅ Batch processing and drag-and-drop folder queueing for high-volume jobs
✅ Custom-model support lets you drop in community-trained Real-ESRGAN weights
✅ Optional Upscayl Cloud tier for users without a capable local GPU

Cons

⚠️ GPU-accelerated but still slow on integrated graphics or older hardware; a 4x pass on a 4K photo can take minutes
⚠️ Quality on faces and fine text still lags dedicated commercial tools like Topaz Gigapixel or Magnific
⚠️ No built-in denoise, face-restoration (GFPGAN/CodeFormer) or generative fill — it's pure super-resolution
⚠️ Occasional hallucinated texture patterns on foliage, hair or heavy JPEG artifacts
⚠️ Cloud tier documentation and model selection is thinner than the desktop app

Use cases

Restoring old family photo scansEnlarging low-res product photography for printUpscaling digital art and illustrationsRetro game sprite and texture enlargementPreparing web images for large-format printBatch enhancement of e-commerce cataloguesOffline privacy-sensitive photo processingCleaning up compressed social-media images