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 — an always-visible run bar with per-image counts, ETA, and progress on the Windows taskbar.
- Pause, resume, and cancel — pause takes effect within seconds; a cancelled run leaves nothing half-written.
- Keeps going past bad images — a corrupt file is reported and skipped instead of stopping the whole batch.
- Input queue — line up multiple folders and images in a single run, and remove any of them individually.
- Keep or replace previous results — send every run to its own timestamped folder, or replace the last run. Replace only ever removes files PixelForge created: anything else in the output folder is never touched, and results never overwrite each other.
- Checks the space first — before a run, PixelForge estimates how much it will write and warns you if the output drive may not have room.
- Recent runs — the dashboard lists your latest runs with what they did, and opens any run's output in one click.
- Recursive scanning — optionally include subfolders and preserve their structure in the output.
- Before / after preview — a comparison slider with zoom, pan and 1:1 actual-pixel view, fullscreen, and next / previous through every result.
- Custom output naming — rename outputs with templates such as
{name},{model},{scale}. - Light and dark themes with a customizable accent colour — text stays readable (WCAG AA) whatever colour you pick.
- Settings save as you go — paths are checked before they are saved, so a mistake is caught in Settings rather than mid-run.
- Keyboard shortcuts for adding inputs, starting a run, and moving between pages.
- Desktop notifications when a batch finishes.
- One-click updates — PixelForge checks GitHub for new releases, verifies each download against its published SHA-256 checksum, and installs it with a single "Restart to update". Updates that can't be verified are refused.
- 100% offline and private — zero telemetry, zero cloud, zero accounts.
| Settings | About |
|---|---|
 |
 |
- Windows 10 / 11 (64-bit)
- Roughly 275 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 |
PixelForge downloads upscayl-ncnn and caesium-clt from their official releases on first launch and runs them unmodified, as separate programs. Full attribution and license notices are shown on the license screen of every installer (build/license.txt).
Requires Node.js 22.12 or newer.
git clone https://github.com/refora-technologies/pixelforge.git
cd pixelforge
npm installnpm install also turns on a pre-commit check (lint, type-check and the fast test suites, a few seconds).
Note:
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 build
# Run the test suites (the pipeline tests use the engine binaries the app installs)
npm test
# Before publishing: walk the installer pages and verify the release through the real updater
npm run verify:installer
npm run verify:release
# Install an older release, then update it the way "Restart to update" does (on a PC without PixelForge installed)
npm run verify:update-install -- -Old path\to\older\PixelForge-Setup.exe
# After publishing: check the live release the way installed copies will see it
npm run verify:published -- --downloadPixelForge is free software, licensed under the GNU General Public License v3.0 or later (GPL-3.0-or-later). © 2026 Refora Technologies.
You may use it for any purpose, including commercial work, share it, and change it. If you share a changed version, it must also be under the GPL, with its source code available.
Versions up to and including 1.2.0 were released under the MIT License.
The open-source components PixelForge uses keep their own licenses (see Open Source Stack and build/license.txt).