Upscale Image 4×
4× in one pass — a 500 px image becomes 2000 px with reconstructed detail. Runs on your device; nothing is uploaded.
Upscale
Cleans JPEG artefacts and noise while keeping detail — the right default.
About the model
Real-ESRGAN "general x4v3" (SRVGGNetCompact, BSD-3-Clause), 4.9 MB, fp32. Runs in a Web Worker with ONNX Runtime on your CPU, in 256-pixel tiles; the 2× option runs the 4× model and box-filters the result down, which is sharper than a native 2× would be. The three styles are the official weights blended with the "denoise" weights at 0, 0.5 and 1. Each style is cached on your device after its first download.
What this tool does
It runs Real-ESRGAN — a compact super-resolution neural network released under the BSD-3 licence — on your device, in a Web Worker with ONNX Runtime, to enlarge the image 4× with reconstructed edges and texture and compression artefacts removed. The 4.9 MB model downloads from this site once and is cached; the photo never leaves your browser. There is no queue, no credits and no watermark because no server does any work.
When 4× is the right choice
Tiny sources: a 200–600 px web image, an old thumbnail, a sprite, a logo saved from a website, an icon you need at poster size. At 4× the model has the most to invent, so results are best on graphics, products, architecture and textures; faces can look over-smooth at 4× — for portraits, 2× is the safer setting. Input is capped at about 1 megapixel (1024×1024) for 4×, which is deliberate: anything bigger does not need quadrupling and the output would be 16 MP.
Speed
4× costs the same as 2× per input pixel (the 2× mode is the 4× model downsampled), so a 512×512 image takes about 5 seconds on a laptop and the page shows a tile counter. Output up to 16 MP as PNG or JPEG.
Limits and honesty
- Super-resolution invents detail. It cannot make unreadable text readable or recover a face that was never captured.
- Inputs are capped at 2.1 MP for 2× and 1.05 MP for 4×; larger images are reduced first and the page says so.
- It is CPU-bound: about 15–40 s for a 1 MP image at 2× on a laptop, longer on a phone, with a progress counter and a cancel button.
The full tool with every option lives at Upscale Image.
Frequently asked questions
Is my photo uploaded?
No. The page is static and has no upload endpoint; your browser does the work on your own device and the result is saved straight back to your disk. The site's Content-Security-Policy only allows it to talk to its own origin, so it cannot send your file anywhere even by accident. Load the page, disconnect from the internet, and it still works.
Why is the input limited to about 1 MP?
Because 4× of anything larger produces a 16+ megapixel image and takes minutes on a CPU — and images that size do not need upscaling. Larger inputs are reduced to the cap first and the page tells you.
4× or 2× for a photo of a person?
2× for faces. 4× works but can look waxy on skin; if you need 4× anyway, use the Smooth style and expect to touch up.
Which model is it?
Real-ESRGAN "general x4v3" (Tencent ARC, BSD-3-Clause) converted to ONNX by us from the official weights, running with ONNX Runtime compiled to WebAssembly. The three styles are the official base/denoise weight blends.
Is it free? Any limits or watermark?
Free, no account, no watermark, no daily quota. Because nothing runs on our servers there is nothing to meter. The only limits are your device's memory — anything up to roughly 50 megapixels is fine on a modern phone or laptop.
Does it work on my phone?
Yes on recent phones; the model download is 4.9 MB and processing takes 30–90 s for a 1 MP image. Keep the tab in the foreground.