Make a small or blurry picture bigger without the usual mushy look. A Real-ESRGAN AI model rebuilds edges and texture right on your device, so the photo is never uploaded to a server.
Drag & drop, click to browse, or paste
JPG, PNG, WebP, AVIF, HEIC and more
Runs fully in your browser. Your files are never uploaded.
Drop, paste or choose a small, blurry or low-resolution image. Images up to about 2 megapixels work best.
Choose 2x or 4x and press Upscale. The AI model downloads once, then your browser enhances the photo tile by tile.
Drag the slider to compare the original with the enhanced version, then save it as PNG, JPG or WebP.
Your files are processed on your device and never uploaded to any server.
No account, no limits and no watermark on your results.
Runs at native speed in your browser and keeps working once the page has loaded.
Get a 2× or 4× image with restored detail. Very large inputs are capped first, and the tool tells you when that happens.
Normal resizing stretches existing pixels, which makes edges soft and blocky. The AI model has learned what sharp edges, skin, text and foliage look like and redraws that detail while it enlarges, so the result looks cleaner at the new size.
Images up to about 2 megapixels, for example 1600 x 1200 px, are processed at their full size. Larger images are reduced to fit before enhancing so the page stays responsive on your device.
The model runs on your processor inside the browser, and the image is processed in small tiles. Expect a couple of seconds for a small icon and roughly one to two minutes for a 2 megapixel photo. A progress bar shows how far along it is and you can cancel at any time.
It uses Real-ESRGAN general x4v3, an open-source model released under the BSD-3 license, which is tuned for photos and illustrations. It is about 5 MB and is downloaded the first time you use the tool.
Yes. The AI enhances the color, and the transparency of the original is enlarged separately and applied again, so a transparent PNG stays transparent when you save as PNG or WebP.
