Real-ESRGAN super-resolution

Upscale an image 2x or 4x

Enlarge a photo without the mush. Super-resolution reconstructs plausible detail instead of stretching the pixels you already had — the difference between a usable print and a blurry one.

2x and 4xReal-ESRGAN modelWorks on cutouts too

How it works

  1. 1

    Upload the image

    JPG, PNG or WebP up to 10 MB. Transparent PNGs work too.

  2. 2

    Pick 2x or 4x

    2x costs 1 credit, 4x costs 2. Both run in the same editor as everything else.

  3. 3

    Download

    A larger image with reconstructed detail rather than stretched pixels.

Why an ordinary resize looks bad

Enlarging an image in a normal editor uses interpolation: to invent a new pixel it averages the ones around it. That is mathematically reasonable and visually terrible, because averaging is exactly what blurring is. Doubling an image's size this way gives you four times the pixels and no extra information.

Super-resolution takes a different approach. A model trained on millions of image pairs has learned what real detail looks like at higher resolution, so it reconstructs plausible texture — edges stay sharp, fabric keeps its weave, text stays legible.

It is reconstruction, not recovery. The model produces detail consistent with what it sees, and it is not reading information that was never captured. For photographs that distinction rarely matters. For anything forensic or evidential, it matters enormously and you should not use an upscaler at all.

When upscaling helps most

  • Old photos and scans that are simply too small for modern screens.
  • A product image you need at banner size but only have as a thumbnail.
  • A logo you were sent as a small raster file with no vector original.
  • Print work, where 300 DPI turns a screen-sized image into a postage stamp.
  • Cutouts you want to use larger than the source photo allowed.

Where it will not save you

Heavy motion blur, severe JPG compression and badly out-of-focus shots do not improve. The model will happily produce a large, sharp-looking version of a blurry image, and the result usually looks synthetic rather than better.

Faces are the case to check. Super-resolution can subtly change facial features when there was not enough detail to work from, which is fine for a background extra and not fine for a portrait of someone you know.

Cutout first, then upscale

If you are doing both, remove the background before upscaling. Working on the smaller file is faster, and upscaling a clean cutout gives the model a clearer subject to reconstruct than upscaling a busy scene and cutting it out afterwards.

Transparent PNGs survive the upscale with their alpha intact, so the order costs you nothing.

Frequently asked questions

What does upscaling cost?
2x costs 1 credit and 4x costs 2. A free account comes with 10 bonus credits, and pay-as-you-go is $2 for 10 credits with no subscription.
Which model do you use?
Real-ESRGAN, a well-established super-resolution model that holds up well on photographs and illustration alike.
Can I upscale a transparent PNG?
Yes. The alpha channel is preserved, so a cutout stays a cutout.
Will it fix a blurry photo?
No. Upscaling adds resolution, not focus. Motion blur and out-of-focus shots come back larger and still blurry. The separate photo-enhance tool does denoise and sharpen, which helps with softness but not with genuine blur.
How large can the output be?
4x from a 10 MB upload, which covers most print and banner needs. Start from the largest original you have for the best result.

Keep exploring

Ready to cut something out?

Three free background removals a day without an account. Sign up free and it becomes five a day plus 10 bonus credits for the premium tools.