AI-powered batch image upscaling and compression for Windows. Free, offline, private.
PixelForge is a Windows desktop app that automates a two-stage image enhancement pipeline:
- AI Upscaling — Upscale images up to 4x using 7 bundled AI models, powered by upscayl-ncnn.
- Smart Compression — Compress the upscaled output using caesium-clt without visible quality loss.
All processing is 100% local. No cloud, no accounts, and no internet required after the one-time dependency setup.
- Pick a folder or individual images — process an entire folder or hand-pick specific images from anywhere, with drag-and-drop support.
- Three pipeline modes — Upscale + Compress, Upscale only, or Compress only.
- 7 bundled AI models — all included in the installer, no separate download needed.
- GPU-accelerated — Vulkan-powered inference via upscayl-ncnn (NVIDIA, AMD, Intel), with automatic priority for the dedicated GPU.
- Batch processing with live progress — exact per-image counters, elapsed time, and ETA.
- Pause, resume, and cancel — stop or hold a batch safely between images.
- Input queue — line up multiple folders and images in a single run.
- Recursive scanning — optionally include subfolders and preserve their structure in the output.
- Before / after preview — inspect results in a gallery with a side-by-side comparison slider.
- Custom output naming — rename outputs with templates such as
{name},{model},{scale}. - Light and dark themes with a customizable accent color.
- Desktop notifications when a batch finishes.
- Built-in update checker — checks GitHub for new releases and downloads them in-app.
- 100% offline and private — zero telemetry, zero cloud, zero accounts.
| Settings | About |
|---|---|
 |
 |
- Windows 10 / 11 (64-bit)
- Roughly 230 MB (includes all 7 AI models)
- No Upscayl installation required
On first launch, PixelForge downloads two small command-line tools (the upscayl engine and Caesium CLT, about 25 MB total) and verifies them before use. The AI models are already bundled with the installer.
| Model | Optimized For |
|---|---|
| Upscayl Standard 4x | General photography |
| Upscayl Lite 4x | Fast processing, lower VRAM |
| Ultra Sharp 4x | Maximum sharpness |
| Remacri 4x | Real-world photos |
| UltraMix Balanced 4x | Balanced output |
| Digital Art 4x | Illustrations and art |
| High Fidelity 4x | High-detail preservation |
PixelForge is an automation layer built on outstanding open-source tools:
| Tool | License | Purpose |
|---|---|---|
| upscayl-ncnn | AGPL-3.0 | AI upscaling engine (Vulkan/NCNN) |
| upscayl-custom-models | MIT | Trained AI model weights |
| caesium-clt | GPL-3.0 | Image compression CLI |
| Electron | MIT | Desktop application framework |
| electron-store | MIT | Persistent settings storage |
| extract-zip | BSD-2 | ZIP extraction for dependency setup |
Full attribution and license texts are included in every installation via the EULA screen.
git clone https://github.com/refora-technologies/pixelforge.git
cd pixelforge
npm installNote:
src/models/is not included in the repository (the files are roughly 180 MB, too large for GitHub). To run in development, copy your AI model files (.paramand.bin) intosrc/models/. Models are available at upscayl-custom-models.
# Run in development mode
npm start
# Build the Windows installer (requires model files in src/models/)
npm run buildMIT (c) 2026 Refora Technologies
This project is MIT licensed. Third-party binaries (upscayl-ncnn, caesium-clt) are distributed under their respective licenses (AGPL-3.0 and GPL-3.0). See LICENSE and the EULA for full attribution.