Upscale Image

Enlarge an image by a factor of two, three or four — cleanly resampled, with nothing invented.

Secure server processingImagesWorkingNo account needed

What this tool does

Upscale Image enlarges a picture by resampling it. The server opens the file with libvips, computes a new pixel grid using a Lanczos3 kernel, and writes the result at two, three or four times the width and height of the picture as it is seen. That distinction matters for a phone photo held upright, which is stored sideways with an EXIF tag telling viewers to turn it: the rotation is applied before the enlargement, so a 3024 × 4032 portrait shot doubles to 6048 × 8064 rather than being squeezed into a landscape frame. Lanczos is the resampler photo editors reach for when enlarging, because it holds edges together far better than the bilinear stretch a browser applies when it draws an image larger than its natural size.

What it will not do is invent detail. Resampling can only redistribute the information already in the file, so a 400 px face at 4x becomes a 1600 px face that is smooth rather than sharp — larger, not clearer. Tools sold as super resolution run a trained model that fills that gap with plausible texture; this does not, which means nothing reaches the output that the camera did not record, and equally that a soft or motion-blurred original stays soft. If reconstruction is what you actually need, this is the wrong tool and no setting here changes that.

The output format is not a choice you make — it follows the source. An image carrying an alpha channel is written as PNG so the transparency survives; everything else is written as JPEG at quality 92, so a screenshot saved as an opaque PNG comes back as a JPG, which is the wrong direction for a text-heavy capture. The enlarged copy also carries none of the original's EXIF block or embedded colour profile, and an animated GIF is opened as a still, so only its first frame is enlarged. The file is renamed with the factor appended, which turns photo.jpg into photo-2x.jpg.

The work runs on the OmniKitly server rather than in the page, because the canvas a browser hands a script has no Lanczos setting to reach for — only a smoothing flag. Your picture is uploaded, enlarged by sharp, and offered back as a download; the copy you sent, the copy you get and the job's own record are then swept off disk together when that job's retention window runs out. One ceiling governs the arithmetic: 100 megapixels, checked against the picture that arrives and again against the size it would become. That leaves roughly 25 MP of source at 2x, 11 MP at 3x and 6 MP at 4x, and a single file over the line stops the whole batch, with both sizes named, so a smaller factor is the way through.

How to use it

  1. Add the images. Anything libvips can decode — JPEG, PNG, WebP, AVIF, TIFF, GIF — plus HEIC straight off an iPhone, which is decoded separately first because the libvips build here carries no HEVC decoder of its own. Several go in one job, up to 50 files with none over 200 MB.
  2. Pick the factor. 2x, 3x or 4x, applied to both sides as the picture is oriented, so a 1200 × 800 photo at 2x arrives as 2400 × 1600 and a portrait one stays portrait. The larger the factor, the softer the result, because there is no extra detail to draw on.
  3. Leave sharpening on, or turn it off. Enlarging softens edges, and the light unsharp mask restores the apparent crispness without adding texture. Switch it off for line art and flat graphics, where sharpening can outline edges you would rather leave clean.
  4. Collect the result. Each file comes back named after its source with the factor appended, as a JPG unless the original had transparency, in which case it is a PNG. Download it before the retention window closes.

Processing & privacy

Secure server processing. This operation requires temporary server processing.

Common use cases

Printing a photo larger than its pixel count allows

A 1500 px image is thin across an A4 sheet. Doubling it gives the printer enough pixels to work with, though the detail remains whatever the camera recorded.

Clearing a minimum-width requirement

Marketplaces, stock libraries and ad platforms reject images under a stated width. Enlarging one that is close gets it accepted without a reshoot.

Making a small logo usable on a large surface

Banners and slides need more pixels than a website logo carries. Where a vector original exists, export that at the size instead — it will be genuinely sharper than any enlargement.

Enlarging a batch of scans on one setting

Add the whole set, choose 2x once, and every file comes back at twice the width and height it is seen at, with the same sharpening applied.

Questions

Is my image uploaded?
Yes — this is one of the few tools here that cannot avoid it, since the resampler lives in libvips rather than in the browser. The picture crosses an encrypted connection to reach it, and everything the job touched is removed once its retention window runs out.
Will it make a blurry photo sharp?
No. This is resampling, not reconstruction, so a soft original becomes a bigger soft image. The sharpening option only restores crispness lost in the resampling step itself; it cannot recover focus the lens never had.
Why did my PNG come back as a JPG?
The output format follows the source's alpha channel rather than its extension. A PNG carrying transparency stays a PNG; an opaque one is written as JPEG at quality 92. On a photograph that is a fair trade for the file size, but on a screenshot or flat artwork JPEG leaves faint ringing around text and hard edges, and there is no format setting here to prevent it.
Why was my 4x job refused?
Both the input and the result are capped at 100 megapixels. A 4000 × 3000 photo at 4x would be 16000 × 12000, or 192 megapixels, so the job stops and names both sizes. Choose 2x, or crop the picture first.
Does the enlarged file keep its EXIF and colour profile?
No. The copy is written fresh, without the camera metadata or the ICC profile the original carried, so a wide-gamut source can look slightly different from the file you started with.
Can I enlarge an animated GIF?
Only its first frame. The file is opened as a still image, so what comes back is one enlarged frame rather than a moving picture.
How is this different from Resize Image?
Resize Image works in your browser and ships with Never enlarge switched on, so asking it for more pixels than the source has holds the image at its original size. This tool is the deliberate exception, with a resampler picked for enlarging — at the cost of an upload.

Related tools