Proxy Scaler
From decklist to
print-ready proxies.
Proxy Scaler fetches Magic: The Gathering card images from Scryfall, runs them through local upscaling models, and lays them out into a print-ready PDF — cut guides, bleed, and double-faced cards all handled for you.
Windows · macOS (Apple Silicon) · Linux — no account, no cloud, runs entirely on your hardware.
Three steps, one PDF
Import a decklist
Paste a decklist using the set / collector-number format — 1 Sol Ring (c21) 263 — so you get the exact printing you want. Archidekt and Moxfield don't automatically export this, please update your Export Options. Generic 1 Sol Ring lines work too, and both styles can be mixed in one list.
Generate upscaled images
Card art is fetched from Scryfall and upscaled up to 1200 DPI with your choice of upscale model. Generation runs in a background queue with live task status.
Export the PDF
Pick a page size, tune bleed, spacing, and cut guides, and download a print-ready PDF. Prefer another layout tool? The raw PNGs are yours too.
Built for the way proxies actually get printed
Upscale models tuned for card art
Three bundled models — UltraSharp v2, IllustrationJanAI, and a fast Real-ESRGAN variant — all trained on illustration rather than photos. Output at 600, 800, or 1200 DPI.
Print-ready PDF layout
Page-size presets or custom dimensions, bleed, spacing, offsets, and cut-guide width and length. It counts remaining print slots and flags any card still missing an image before you print.
Double-faced cards, handled
Transform and modal double-faced cards get both faces fetched, upscaled, and printed. Split, flip, and adventure cards come through as the single face they are.
Private, local, and free
Everything runs on your machine — no account, no uploads, no subscription. Works on CPU out of the box; a GPU just makes it faster.
Every GPU welcome
Nvidia via CUDA, Apple Silicon via Metal, AMD via ROCm on Linux and DirectML on Windows. Dedicated builds per platform, no toolkit installs.
Use a beefier box
The desktop app can hand generation to a server running on another machine — say a GPU desktop reached over your LAN or Tailscale — while your projects stay local.
A proper desktop app
No save button, nothing to configure first — paste a decklist and go. Everything persists as you work.
Download
Installers are self-contained — Python, PyTorch, and the model loader are all bundled. Pick your OS, and on Windows the build that matches your GPU. Current version: 0.1.0.
Windows
Nvidia RTX 20-series or newer, GTX 16-series — also the right pick with no GPU at all cuda
Nvidia GTX 10-series or older — don't use on RTX 50 cards cuda-legacy
AMD or Intel — any DirectX 12 GPU directml
macOS
Apple Silicon (M-series) — Metal acceleration built in macos-arm64
Linux
Standard build for Nvidia or CPU-only, rocm build for AMD — plus a headless Debian/Ubuntu server package
The desktop app is what most people want; the server app hosts generation for other machines to connect to. Prefer the terminal? A Python CLI and a headless Debian/Ubuntu server package are covered in the README.
Help shape what's next
Back printing, custom image upload, and MPCFill-style art browsing are on the roadmap — the order is up for debate, and that debate happens on Discord.