AI image upscaler

Small photo, big format

An old phone photo, an image saved from the web, or a logo that only exists as a thumbnail: when you need them bigger, plain stretching blurs them. Here a neural network upscales them and fills in edges and fine detail, so the result holds up in print, a presentation or a shop listing. See the difference immediately on the comparison slider next to the original.

And the important part: unlike most “AI upscalers”, your photo is never uploaded. The model downloads into your browser (about 5 MB, first use only) and all the computing happens on your machine. Family photos and work materials never leave your computer.

Two modes

AI mode is best for photos, drawings and logos up to 4 megapixels: it reconstructs detail, so hair, grass and lettering stay crisp. It is fastest in browsers with WebGPU (Chrome, Edge); without it, it takes longer. Fast mode is classic sharpened resampling: instant, works up to 40 megapixels and is plenty for screenshots or large images.

Frequently asked questions

Will AI make a blurry photo sharp?

No miracles. The AI adds detail that plausibly belongs in the source, so a small but clean image comes out great. A heavily blurred or very noisy source improves only partially.

Is it really free and upload-free?

Yes. The tool runs entirely in your browser, so an upscale costs us nothing and your photo stays with you. No sign-up, no watermark.

Which model does it use?

Real-ESRGAN (the realesr-general-x4v3 variant), an open model under the BSD licence by Xintao Wang et al., running on ONNX Runtime. Scales below 4× are downsampled from the network output, which is the standard approach.

Does it handle transparent PNGs?

Yes, transparency is preserved.