Skip to content

AI Image Upscaler: 2× and 4×

Enlarge a photo to 2× or 4× its size with an AI super-resolution model that runs in your browser, sharper than plain resizing. Save as PNG or JPG.

Private · on device. Runs in your browser: your file is not uploaded. No sign-up.

JPG, PNG, WebP, HEIC (iPhone), BMP or AVIF

What an AI upscaler does

A small picture has a fixed number of pixels. Blowing it up with ordinary resizing only stretches them, which is why enlarged photos look soft and text goes fuzzy. A super-resolution model is a neural network trained on millions of pairs of small and large pictures, so it predicts the missing detail: sharper edges, cleaner lettering, finer texture.

This tool uses Swin2SR, an open model released under the Apache-2.0 licence. The 2× model, a lightweight version that is quick enough for a browser, doubles the width and height (four times the pixels); the 4× model, trained on real-world photos, multiplies each side by four (sixteen times the pixels).

How it works in your browser

  1. The model is downloaded once from Hugging Face and kept by your browser.
  2. Your picture is cut into tiles with a little overlap, so nothing is lost at the joins.
  3. Each tile is upscaled on your processor, and the tiles are stitched back together.
  4. The result is saved as PNG or JPG. Nothing is uploaded at any point.

Getting the best result

  • Start from the original file, not a screenshot of it or a copy sent through a chat app, which is compressed.
  • Use 2× for logos, screenshots and text; 4× for small real photos.
  • Need an exact size afterwards? Use the resize tool; to bring the file under a limit, the compressor.

Honest limits

An upscaler invents plausible detail; it does not reveal real detail that the camera never captured. Faces in tiny photos can come out subtly different, and text that is unreadable to start with will not become readable. Use it to make a picture look better, not as evidence of what was there.

Frequently asked questions

How is this different from just resizing the picture?

Ordinary resizing spreads the existing pixels over a bigger area, so edges go soft and blurry. A super-resolution model has learned what sharp edges, text and textures usually look like, and fills in the new pixels accordingly. The result is crisper, though it cannot recover detail that was never captured.

Is my picture uploaded?

No. The AI model (about 7 MB for 2×, 21 MB for 4×) is downloaded to your browser once and runs on your device. The picture itself never leaves it.

Why does it take so long?

Every output pixel is computed by a neural network running on your processor, which is heavy work for a browser. The tool shows an estimate before it starts and the time left as it goes. On a typical computer a 1000 × 1000 px picture takes about four minutes at 2×; 4× is about five times slower, and phones are slower still, so on a phone start with 2× and a small picture.

Is there a size limit?

Yes, so the work finishes in reasonable time and the browser tab does not run out of memory: on a computer the result is kept to about 8 megapixels at 2× and 4 megapixels at 4×, and half that on a phone. A bigger picture is reduced first, and the tool says so. Upscaling is most useful for small pictures anyway: old photos, thumbnails, screenshots and logos.

Can it fix a blurry or out-of-focus photo?

Only a little. The 4× model is trained on real photos and copes with some blur and compression noise, but a badly blurred photo stays blurred, only larger. It works best on pictures that are sharp but small.

PNG or JPG?

PNG is lossless and keeps transparency, but files are large. JPG is a fraction of the size and fine for photos you will share or print. Pictures with a transparent background keep it when saved as PNG.