Upscale an image without uploading it
Old photos, WhatsApp pictures and images saved from the web are often too small: enlarge them and they turn into a blurry mess. An upscaler rebuilds the missing pixels instead of stretching the ones you have, so edges stay sharp and faces keep their shape.
This one runs entirely in your browser. The model is downloaded once (4.6 MB) and your picture is processed on your own device, so your photos are never uploaded and nothing is stored anywhere.
How it works
- Add a picture: drop it here, click to choose, or paste a screenshot.
- Choose 2× or 4×, and JPG or PNG.
- Press Upscale. The first run also downloads the model, so it takes a few seconds longer.
- Download the result, or all of them as a ZIP.
What it is good at, and what it is not
It works well on soft photos, small product pictures, screenshots and pictures of text. It gives plausible detail rather than real detail, so a very small face gains sharpness but not identity, and heavy hair or fine fabric can look a little painted. For printing, upscale first and check the result at full size before you send it.
Frequently asked questions
Does upscaling really add detail?
It adds plausible detail, not real detail. The model has seen millions of photos, so it knows what sharp edges, skin and text usually look like and rebuilds them. That is why an upscaled photo looks far better than a stretched one, but a tiny face will not become a portrait.
Are my photos uploaded?
No. The model is downloaded to your browser and your photo is processed on your device. Nothing leaves your computer or phone.
Why does it take so long?
Your own device does the work and every pixel passes through a neural network. A small picture takes a few seconds to about a minute. Very large images are refused, because they are already big enough to use.
2× or 4×: which should I choose?
4× for very small pictures such as an old avatar or a thumbnail, 2× for photos that are nearly big enough and only look a little soft.
Other image tools
Upscaling by Real-ESRGAN (general x4 v3 model, BSD-3-Clause), running in your browser with ONNX Runtime Web.