WarpClyr — Local AI Image Upscaling with Real-ESRGAN

Overview

WarpClyr is a cross-platform desktop application that brings AI-powered image upscaling directly to the user’s machine—no cloud, no subscriptions, no data leaving the device. Built with Electron, React and TypeScript, the app wraps the Real-ESRGAN ncnn-vulkan engine into an intuitive drag-and-drop interface with real-time before/after comparison.

Challenge

Most AI upscaling tools either run in the cloud—raising privacy and cost concerns—or require command-line expertise that excludes non-technical users. The challenge was to package a GPU-accelerated inference pipeline into a polished desktop experience that works across macOS, Windows and Linux without requiring any setup from the user.

Goals

The project was driven by the following goals:

• Deliver a fully local, privacy-first upscaling workflow
• Support multiple Real-ESRGAN models for photos and illustrations
• Provide an interactive compare viewer with zoom presets for quality assessment
• Ship as a single installable package for all three major platforms
• Keep the interface minimal and focused on fast, frictionless processing

Solution

WarpClyr was built as an Electron application with a React 19 frontend, bundled via Vite for fast development and optimized builds. The app ships with pre-bundled Real-ESRGAN ncnn-vulkan binaries and model files for each target platform, eliminating any external dependency.

Users import images via drag-and-drop or native file picker, choose between a photo or illustration model, set the desired scale factor (2x–4x) and output format (PNG or JPG), and start the upscale. A built-in compare viewer with a draggable divider and zoom presets (Fit, 100%, 200%) allows immediate quality assessment side by side. The app persists user settings between sessions and provides clear run-state feedback throughout the process.

Packaging is handled by electron-builder, producing DMG and ZIP for macOS, NSIS and portable builds for Windows, and AppImage and DEB for Linux.

Outcome

The result is a lightweight, self-contained upscaling tool that makes AI image enhancement accessible to anyone—designers, photographers and content creators—without technical barriers. Processing runs entirely on the local GPU, keeping images private and latency minimal.

Learnings

Shipping native binaries across three platforms introduced significant build complexity, especially around architecture-specific asset management and executable permissions. The project reinforced the importance of a robust asset pipeline and platform-aware packaging configuration. Designing the compare viewer also provided valuable insights into performant canvas rendering and smooth interaction patterns in Electron.