bg2x
Browser-based AI image upscaler with 10 local models and zero server uploads.
Upscale photos and art up to 4x using WebGPU and ONNX Runtime entirely in your browser.
bg2x is a free, privacy-first image super-resolution tool running entirely client-side through ONNX Runtime Web. It is built for graphic designers and artists who need to upscale images locally without sending files to external servers.
What it is
The application applies 10 specialized models—including Real-ESRGAN, Swin2SR, and Real-CUGAN—to photos, anime, and compressed images. It supports batch processing, drag-and-drop input, and exports to PNG, JPG, or WebP formats.
Who it's for
It suits designers who handle sensitive imagery and want offline-capable upscaling without cloud subscription fees. It does not suit teams requiring centralized server-side automation pipelines or non-technical users who cannot manage local model downloads.
Key features
Privacy-First Processing
Runs 100% client-side via Web Workers with no backend data uploads.
Multiple AI Models
Includes 10 specialized models for photos, anime, and compressed JPEGs.
GPU Acceleration
Leverages WebGPU when available with an automatic fallback to WASM.
Batch Queue Support
Queue multiple images and download the results as a single ZIP archive.
Up and running in 3 steps
Our honest verdict
What's great
- Completely client-side execution ensures your design files never leave your machine
- Supports hardware acceleration via WebGPU with automatic WASM fallback
- Provides native Electron desktop apps for Windows, macOS, and Linux
- Caches models in IndexedDB after the first download for offline use
Where it falls short
- Requires cross-origin isolation headers on production hosts to enable threaded WASM inference
- Initial model files must be downloaded on demand from Hugging Face or a mirror
- Heavy models like Swin2SR Classical x4 require significant browser memory
Pricing & license
Free under its license, with the only cost being your hardware and bandwidth to download model files.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Are my images uploaded to a cloud server?
No. All inference happens locally in your browser using ONNX Runtime Web inside a Web Worker.
Which AI models are included?
It features 10 models across Real-ESRGAN, Swin2SR, and Real-CUGAN families for photos, anime, and compressed images.
Can I run this as a desktop application?
Yes, native Electron builds are available for Windows, macOS, and Linux.
