Is this really fully in-browser?

Yes. The only network call is the one-time model download. Image bytes never leave your device.

4× costs ~4× the compute of 2× and is rough on laptops / mobiles. We cap Free at 2× so the tool stays usable for everyone.

With WebGPU, a 1024² image at 2× runs in 10-20 s. WASM-only: 1-2 min. First run adds a ~70 MB model download.

8 MB per image. Resize larger originals first — tile inference needs memory overhead.

ONNX Runtime Web + WebGPU 2× / 4× super-resolution

AI Upscale (Real-ESRGAN) — 100% in-browser

Super-resolution without uploading your image. WebGPU on supported browsers; WASM fallback everywhere else.

Real-ESRGAN super-resolution running fully in the browser via onnxruntime-web. Free gets 2×, Pro unlocks 4× — no image ever leaves your device.

Is this really fully in-browser?: Yes. The only network call is the one-time model download. Image bytes never leave your device.
Why is 4× Pro-only?: 4× costs ~4× the compute of 2× and is rough on laptops / mobiles. We cap Free at 2× so the tool stays usable for everyone.
How fast is it?: With WebGPU, a 1024² image at 2× runs in 10-20 s. WASM-only: 1-2 min. First run adds a ~70 MB model download.
Input size limit?: 8 MB per image. Resize larger originals first — tile inference needs memory overhead.