Enlarging is the one image tool that has to invent something. Everything else moves pixels around or spends fewer bytes on them; upscaling has to produce pixels that were never recorded. This is what the tool actually does, what it cannot do, and the one number on the screen that is worth reading before you press the button.
Open the Upscale Image tool — free, no sign-up, no watermark.
What enlarging can and cannot do
A 400-pixel-wide photograph contains 400 pixels of information. Any bigger version has to fill the gaps by reasoning about the neighbours, and what separates a good enlargement from a bad one is entirely how that reasoning is done.
The common methods are crude. Nearest-neighbour repeats pixels and gives you visible blocks. Bilinear averages the four nearest and gives you mush. This tool uses Lanczos, which weighs a wider neighbourhood with a windowed sinc function — it holds edges together and keeps fine texture from dissolving, which is why it is the default in most serious image software.
It is not AI. There is no model here reconstructing what skin or brickwork ought to look like, and the screen offers no model to choose because there is none to choose from. That distinction matters in both directions: a learned upscaler can hallucinate plausible detail into a blurry photo, and this will not. What comes back is a larger, cleaner rendering of exactly the information your file already holds.
Step 1: add your image
The file is read in your browser first, so its size is known before anything is sent.
Accepted formats are JPG, PNG, WebP, AVIF, GIF and TIFF, up to 25 MB.
Step 2: choose how much bigger
Upscale Factor
Each tile carries the dimensions that factor produces for the picture you loaded.
Three fixed factors and a custom option, with the proportions always kept. Start at 2× and look at the result before reaching for more: quality falls away as the factor rises, because more of the final picture is inferred rather than recorded. At 2× three-quarters of the pixels are new. At 4×, fifteen-sixteenths. At 8×, sixty-three sixty-fourths.
The number worth reading
On a large photograph, 4× and 8× are the same answer — and the tiles say so.
Results are held to 8000 pixels on the longest side. Rather than print a size it cannot deliver, a tile that runs into that ceiling shows what you will really get and what factor that works out at.
A 3000 × 2000 photograph asked for 4× comes back 8000 × 5333 — which is 2.7×, not the 12000 × 8000 the arithmetic implies. Ask for 8× and you get exactly the same file. That is worth seeing before you choose, not after. On a picture already at 8000 px every tile is disabled and the button is off, because there is nothing to be gained by pressing it.
Custom size
An exact width and height instead of a factor. Keep ratio derives the second number from the first so the picture is never stretched. Use this whenever you know the number you need: going straight to 3000 px is better than taking 4× and landing on 3200, because fewer pixels were invented to reach the same finished size.
Estimated Result
The dimensions here are exact — they are the server's own arithmetic. The file size is deliberately a range, because how much an enlarged file weighs depends on the picture: an enlargement is smooth where the original was detailed, so it compresses better per pixel than the pixel count alone would suggest.
Measured across smooth, noisy, high-contrast and photographic images, the bytes tracked the pixel ratio raised to a power between 0.64 and 0.86. At 4×, that puts a 500 KB file anywhere between about 2.9 MB and 5.3 MB depending on content. A single number there would be a guess wearing a suit.
Step 3: compare, then download
Drag the handle, or focus it and use the arrow keys.
The finished screen puts the original and the enlargement under one divider, each labelled with its pixel size. Use it every time. Enlargement is the one operation where the honest answer is sometimes "this did not help" — and the wipe is where you find that out, rather than after you have already used the file.
Where the honest edges are
Enlarging will not fix a blurry photo. Blur is missing information and there is nothing to reconstruct it from, so you get a larger blurry photo. The same goes for camera shake, heavy JPEG artefacts and noise, all of which are magnified along with everything else. A heavily compressed source enlarges badly for exactly this reason.
There is no denoising or sharpening pass, so a grainy original comes back grainy at a larger size. Every result is capped at 8000 pixels per side. Images go through one at a time, up to 25 MB, and iPhone HEIC files are refused. The enlargement is computed on a server rather than in your browser, so the file does travel; it is processed and returned straight away, never written to disk and never shared.
Common questions
How do I make an image bigger without losing quality?
Strictly, you cannot — enlarging always means inventing pixels, so the result is softer than a photograph taken at that size. What you can do is invent them well. Lanczos resampling weighs a wide neighbourhood rather than averaging the nearest few, and holds edges and texture together far better than the scaling built into a browser or an office suite. Start at 2×, and check the comparison wipe.
Why does 8× give me the same file as 4×?
Because both ran into the 8000-pixel ceiling. On a 3000 × 2000 photograph, 4× would be 12000 × 8000 and 8× would be 24000 × 16000 — both past the limit, so both return 8000 × 5333. The tiles show that size and label it "capped", so the two are visibly the same answer before you press anything.
Does this use AI to upscale?
No. It is a Lanczos resize — a considerable step up from bilinear scaling, but not a model that has learned what things ought to look like. A learned upscaler can hallucinate plausible detail into a blurry photo; this will not.
What is the largest image I can produce?
8000 pixels on the longest side, roughly a 42-megapixel picture at 4:3. A picture already that large cannot be enlarged here at all, and the screen disables every tile and the button rather than returning your file unchanged and calling it a result.
Should I use a factor or a custom size?
Use Custom whenever you know the number you need. The factors are for when "bigger" is the requirement and the exact figure does not matter.
How big will the enlarged file be?
Bigger, but not in proportion to the pixel count. Measured across four kinds of image, the bytes tracked the pixel ratio raised to a power between 0.64 and 0.86 — which is why the screen prints a range rather than a single number.
Does upscaling remove EXIF and GPS data?
Yes. Camera make and model, the date and any location tags are stripped from the file you download. The rotation tag is read and applied to the pixels first, though, so a phone photo comes back upright rather than on its side.
What is the difference between upscaling and resizing?
Resizing goes both ways and is usually used to make a picture smaller, which throws information away safely. Upscaling only goes up, and has to invent what it adds. If you are making something smaller, use Resize Image — it is the same operation without the guesswork.
Try the Upscale Image tool, or browse the rest of the Image Toolkit.