Enlarge a photo without the blur
Drop in a small, old or low-resolution photo and get it back four or eight times larger. An AI model rebuilds edges and texture instead of simply stretching the pixels, so the result stays sharp. Everything runs on your own device — the photo never leaves your browser.
How to enlarge a photo without it going soft
Enlarging a photo the ordinary way is a form of guessing: the software takes the pixels it has and spreads them over a bigger canvas, filling the gaps with an average of the neighbours. That is why a stretched photo looks soft — no new detail is created, the same detail is simply smeared over more pixels. An AI upscaler works the other way around. It has seen millions of pairs of small and large images, so it can tell what a sharp edge, a strand of hair or a brick wall should look like at four times the size, and draws that instead of a blur.
What it does well, and what it cannot do
This works best on photos that are simply small: an old scan, a picture from a website, a profile photo, a product shot you only have at thumbnail size. Edges get crisp again, text on a sign becomes readable, skin and fabric keep their texture. What it cannot do is invent information that was never captured. A face that is twelve pixels wide will come out as a sharper twelve-pixel face, not as a portrait — and a photo that is out of focus or badly motion-blurred stays out of focus, because the model cannot know what was behind the blur.
4x or 8x, and how long it takes
Four times is the everyday setting: a small photo comes back at four times the width and height, with edges and texture rebuilt instead of stretched. Eight times goes further and is meant for the really tiny originals — an old thumbnail, a logo pulled off a website, a crop from a bigger picture. The bigger the enlargement, the more the model has to fill in, so on a photo that is already sharp the extra step adds little; on a small one it makes the difference between usable and not. Your photo is worked through in tiles of 128 pixels that overlap on purpose, so the seams can never show up as a grid in the result.
Nothing is uploaded
The whole thing runs inside your browser: the model is downloaded once and the photo is processed on your own processor or graphics card. Your photo is never sent to us or to anyone else, which matters for family pictures, ID scans and anything you would not want on a stranger's server. It also means the speed depends on your device — a modern laptop with a graphics card finishes a one-megapixel photo in seconds, an older phone will need a bit of patience.
About this tool
Real-ESRGAN, the same family of models the desktop upscalers use, running locally in your browser. Free for photos up to two megapixels at 4x, with no watermark on the result.
Why is my big photo not free?
Enlarging costs real computing time, and it grows with the number of pixels: a four-megapixel photo is four times the work of a one-megapixel one, and the 8x setting runs the model a second time on an image that is already sixteen times bigger. The free tier covers the everyday case — a small or old photo at 4x, without a watermark. Bigger photos and the 8x setting are part of the paid tier, and an Images & media plan or day pass covers them, as do credits.
Will it fix a blurry or out-of-focus photo?
No, and it is worth being clear about that. This model makes a small photo bigger while keeping it sharp; it does not recover detail that the camera never recorded. If the photo is out of focus or the subject moved, the result is a larger version of the same blur. Where it does help is a photo that is sharp but simply too small.
Is my photo uploaded anywhere?
No. The model runs in your browser and the photo stays on your device. You can check it yourself: open your browser's network tab and you will see the model being downloaded, but nothing being sent back.
Which formats can I put in and get out?
In: anything your browser can read — JPG, PNG, WebP, GIF, BMP, and HEIC on Apple devices. Out: PNG (lossless, largest file), JPG (smallest) or WebP (a good middle ground). PNG is the default because it keeps every bit of the detail the model just rebuilt.
Why does it take longer on my phone?
Because your phone is doing the work. The photo is processed in tiles and every tile goes through the model, so the time scales with the number of pixels and with how fast your device is. On a computer with a graphics card the browser can use it and the job is much quicker; on a phone the processor does it alone. You can stop a run at any time.
My phone photo is over the 2-megapixel free limit — can I still upscale it for free?
Phone cameras shoot at 10–12 megapixels, well above the free limit — but a photo that large rarely needs upscaling; the ones that do are usually small crops, downloads and AI images. Two practical routes: take a screenshot of the photo on your phone — a screenshot is only as large as your screen, which usually lands it comfortably under 2 megapixels — or scale the photo down first on your phone or computer and upload that version. Anything under the limit upscales for free at 4x.
