The definitive AI-powered image upscaler for 2025.
Most desktop upscalers are built on outdated technology—heavy Electron frameworks consuming hundreds of megabytes of RAM, clunky CLI wrappers, and rigid model support. RustScale breaks this mold:
| Legacy Apps | RustScale |
|---|---|
| 🐌 ~400MB RAM (Electron) | ⚡ ~30MB RAM (Tauri v2) |
| 🔧 CLI wrapper processes | 🧠 In-process AI runtime |
| 📦 Proprietary formats | 🌐 Industry-standard ONNX |
| 🎲 Manual configuration | 🤖 Auto-detects hardware |
| 🔲 Visible grid artifacts | ✨ Seamless tiling |
|
Drag and drop any ONNX super-resolution model—no compilation needed. Automatically detects and optimizes for GPU, NPU, or CPU. Queue multiple images or entire folders with real-time progress. |
Compare before/after with pan, zoom, and slider controls. Optional facial enhancement using CodeFormer integration. Smart defaults that just work, with advanced controls when needed. |
This project was built using AI coding agents: Google Antigravity, Gemini 3, and Claude Sonnet 4.5.
I don't have extensive coding experience, but I had a clear vision for an upscaler that respects both user hardware and modern AI capabilities. Agentic AI coding tools helped me turn that vision into reality—an app I genuinely want to use.
If you have reservations about AI-assisted development, that's completely valid, and you're free not to use this app. I believe in transparency about the tools used in creation. What matters most is whether the end result serves users well.
I'm always open to feedback, suggestions, and contributions that can make RustScale even better. 💡
RustScale uses a "Native Integration Architecture" instead of spawning external processes:
┌─────────────┐ ┌──────────────┐ ┌─────────────┐
│ Svelte │ ───> │ Rust Core │ ───> │ ONNX Runtime│
│ UI Layer │ IPC │ + HAL │ Zero │ (In-Memory)│
└─────────────┘ └──────────────┘ Copy └─────────────┘
│
┌───────┴───────┐
│ GPU/NPU/CPU │
└───────────────┘
Benefits:
- ⚡ Zero-latency communication between UI and AI
- 💾 Direct VRAM management
- 📊 Real-time progress tracking
- ⏹️ Instant cancellation
| Model | Speed | Quality | Best For |
|---|---|---|---|
| 🏃 SPAN | ⚡⚡⚡ | ⭐⭐⭐ | Fast upscaling for low-res images |
| 🚀 SAFMN-V3 | ⚡⚡⚡ | ⭐⭐⭐ | Next-gen efficient model |
| 🎨 NomosUni SPAN | ⚡⚡ | ⭐⭐⭐ | Digital art and illustrations |
| 💎 Real-PLSKR | ⚡ | ⭐⭐⭐⭐ | Maximum quality for photos |
| 🎭 RealPLSKR Anime | ⚡ | ⭐⭐⭐⭐ | Best quality for anime & digital art |
| 🔧 Real-ESRGAN-V3 | ⚡⚡ | ⭐⭐ | Legacy model for compatibility |
| 🪟 Windows | 🍎 macOS | 🐧 Linux |
| Windows 10/11 DirectML support |
macOS 13+ Apple Silicon or Intel |
Ubuntu 20.04+ or equivalent |
Installation Files:
- 🪟 Windows:
.exeinstaller - 🐧 Linux:
.AppImage
- Launch RustScale
- Select your scenario (Fast, Balanced, Quality, or Digital Art)
- Drag & drop images or folders
- Adjust settings if needed (repair strength, face recovery)
- Click upscale and watch the magic happen ✨
| Scenario | Tip |
|---|---|
| 💻 Laptop users | App automatically enables Eco Mode on battery power |
| 🎮 Limited VRAM | Engine adjusts tile sizes based on available memory |
| 🚀 Batch processing | Higher batch sizes improve throughput on powerful GPUs |
Contributions are welcome! Whether it's:
- 🐛 Bug reports
- 💡 Feature requests
- 🔧 Code improvements
- 📖 Documentation
I'd love to hear from you!
This project is licensed under the MIT License - see the LICENSE file for details.
Built with modern, powerful tools:
- 🦀 Tauri v2 - Lightweight desktop framework
- 🧠 ONNX Runtime - Cross-platform AI inference
- ⚡ Svelte 5 - Reactive UI framework
- 🎨 TailwindCSS v4 - Utility-first styling
- 🤖 AI Coding Agents: Google Antigravity, Gemini 3, Claude Sonnet 4.5
