Upscayl — Free Open-Source AI Image Upscaler

Introduction

Upscayl is a free, open-source desktop application that upscales images using AI super-resolution models. It runs entirely on your local machine using Vulkan-based GPU acceleration, so your images never leave your device. Upscayl supports Linux, macOS, and Windows.

What Upscayl Does

• Upscales images by 2x, 3x, or 4x using AI super-resolution models
• Processes single images or entire folders in batch mode
• Runs inference locally using Real-ESRGAN and other NCNN-based models
• Provides a before-and-after comparison slider for quality review
• Exports results in PNG, JPG, or WebP formats

Architecture Overview

Upscayl is built with Electron for the desktop shell and uses Real-ESRGAN-ncnn-vulkan as the inference backend. The Vulkan-based compute pipeline runs on most GPUs across vendors (NVIDIA, AMD, Intel) without requiring CUDA. The application spawns the upscaling binary as a child process, streaming progress updates back to the UI.

Self-Hosting & Configuration

• Install from Flatpak, Homebrew, or the GitHub Releases page
• Select from built-in models or load custom NCNN models
• Configure output format, scale factor, and compression quality
• Set a default output directory for processed images
• Adjust GPU thread count for performance tuning on multi-GPU systems

Key Features

• No cloud dependency — all processing happens on-device
• Vulkan backend works across NVIDIA, AMD, and Intel GPUs
• Batch upscaling for processing entire directories at once
• Support for custom models in NCNN format
• Built-in comparison view to inspect upscale quality

Comparison with Similar Tools

• Topaz Gigapixel AI — commercial and paid; Upscayl is completely free and open source
• waifu2x — older model architecture; Upscayl uses newer Real-ESRGAN models with better results
• Real-ESRGAN CLI — command-line only; Upscayl wraps it in a user-friendly desktop GUI
• ChaiNNer — node-based image processing; Upscayl is simpler for one-click upscaling

FAQ

Q: Does Upscayl require a dedicated GPU? A: A Vulkan-compatible GPU is strongly recommended. CPU-only mode exists but is very slow.

Q: What image formats does Upscayl support? A: Input supports PNG, JPG, WebP, and AVIF. Output can be PNG, JPG, or WebP.

Q: Can I add custom AI models? A: Yes. Place NCNN-format model files in the custom models directory and they appear in the model selector.

Q: How large can input images be? A: It depends on GPU VRAM. For images larger than 4K, Upscayl automatically tiles the image to fit available memory.

Sources

• https://github.com/upscayl/upscayl
• https://upscayl.org