Why does it come out sharper and not just bigger?

Because the new pixels are not copies of the old ones. A plain resize has four times the area to fill and no new information to fill it with, so it spreads each pixel outwards and blends between them, and blending is what soft means. This asks a question a resize never asks: given this small picture, what would the large one have looked like? It has been shown enough pairs of small and large pictures to answer plausibly, so where a resize draws a gradient it draws an edge, and where a resize draws a flat patch it draws grain.

So is the detail real?

It is plausible rather than recovered. Nothing can recover what is not in the file, and a small picture does not contain the large one. What you get is detail that fits: bark that looks like bark, fabric that looks woven, a leaf edge where the blur used to be. On ordinary photographs that is indistinguishable from the real thing in every way that matters, and it is always better than mush. On a face a few dozen pixels across it is exactly as unreliable as it sounds, which is why the answer below about small faces is worth reading before you use one of these on a person.

Can I keep the original size?

Yes, there is a switch above the button. It still enlarges by four on the inside, because that is the only thing the engine knows how to do, and then draws the answer back down to the size you started with. The wait is the same and what you get is a cleaner version of the picture rather than a bigger one. One warning: it does not simply undo itself. What comes back is crisper than what went in, sometimes crisper than the photograph ever was, because making things crisp is what the enlarging was doing all along.

Is my picture uploaded?

No. The enlarging happens on this page, on your own machine, and no request carries the picture anywhere. The one thing that is fetched is the enlarger itself, about 5 MB, the first time you use the tool. Your browser keeps it after that.

Why is there a size limit?

Because your machine is doing the work rather than a rented graphics card. A million pixels is eighty one squares and a couple of minutes; twice that is twice the wait and a result some phones cannot hold. Saying no at the start beats a progress bar that dies at ninety percent.

Small faces, where it will confidently invent features nobody had. Text, which comes back crisp and occasionally wrong. And anything already sharp: it fills in what is missing, so a picture that is missing nothing comes back much the same, only larger.

Will it fix a blurry photo?

Sometimes, and not the way people hope. Softness from having too few pixels is exactly what this is for. Motion blur, a missed focus or heavy compression are damage in the pixels that are there, and enlarging them enlarges the damage too.

Does it work on logos and flat artwork?

It was built for photographs, so that is where it is strongest. Flat art comes back cleaner than a plain resize but can lose the exactness of a hard edge. If you still have the original as an SVG, scaling that will always beat enlarging a picture of it.

ON YOUR DEVICE · FREE

Upscale Image

Four times bigger, and sharp instead of soft

…

1
Choose a picture
Up to a million pixels, which is 1000 by 1000 or any other shape with as many. Small is the point here: this exists for the picture that is already too small, and it is read from disk rather than sent anywhere.
2
It is enlarged a square at a time
The picture is cut into small squares and each one is enlarged on its own, because the thing doing the enlarging works at one fixed size. Every square reads a strip of its neighbours before it starts and is faded into them where they meet, which is what keeps the joins from showing. Budget a second or two per square and watch the counter.
3
Download
Four times the width and four times the height as a PNG, or the size you started with if you asked for that instead. Transparency survives, and nothing of ours is stamped on it.

FAQ

Why does it come out sharper and not just bigger?: Because the new pixels are not copies of the old ones. A plain resize has four times the area to fill and no new information to fill it with, so it spreads each pixel outwards and blends between them, and blending is what soft means. This asks a question a resize never asks: given this small picture, what would the large one have looked like? It has been shown enough pairs of small and large pictures to answer plausibly, so where a resize draws a gradient it draws an edge, and where a resize draws a flat patch it draws grain.
So is the detail real?: It is plausible rather than recovered. Nothing can recover what is not in the file, and a small picture does not contain the large one. What you get is detail that fits: bark that looks like bark, fabric that looks woven, a leaf edge where the blur used to be. On ordinary photographs that is indistinguishable from the real thing in every way that matters, and it is always better than mush. On a face a few dozen pixels across it is exactly as unreliable as it sounds, which is why the answer below about small faces is worth reading before you use one of these on a person.
Can I keep the original size?: Yes, there is a switch above the button. It still enlarges by four on the inside, because that is the only thing the engine knows how to do, and then draws the answer back down to the size you started with. The wait is the same and what you get is a cleaner version of the picture rather than a bigger one. One warning: it does not simply undo itself. What comes back is crisper than what went in, sometimes crisper than the photograph ever was, because making things crisp is what the enlarging was doing all along.
Is my picture uploaded?: No. The enlarging happens on this page, on your own machine, and no request carries the picture anywhere. The one thing that is fetched is the enlarger itself, about 5 MB, the first time you use the tool. Your browser keeps it after that.
Why is there a size limit?: Because your machine is doing the work rather than a rented graphics card. A million pixels is eighty one squares and a couple of minutes; twice that is twice the wait and a result some phones cannot hold. Saying no at the start beats a progress bar that dies at ninety percent.
What is it bad at?: Small faces, where it will confidently invent features nobody had. Text, which comes back crisp and occasionally wrong. And anything already sharp: it fills in what is missing, so a picture that is missing nothing comes back much the same, only larger.
Will it fix a blurry photo?: Sometimes, and not the way people hope. Softness from having too few pixels is exactly what this is for. Motion blur, a missed focus or heavy compression are damage in the pixels that are there, and enlarging them enlarges the damage too.
Does it work on logos and flat artwork?: It was built for photographs, so that is where it is strongest. Flat art comes back cleaner than a plain resize but can lose the exactness of a hard edge. If you still have the original as an SVG, scaling that will always beat enlarging a picture of it.

Upscale Image

Choose a picture

It is enlarged a square at a time

Download

FAQ

More tools like this one