Increase image resolution, free
Drop in an image and an AI model enlarges it two or four times — sharper than a plain resize. The picture stays on your device the whole time.
Drop an image here, or choose a file
JPG, PNG or WebP, up to 16 megapixels. The image stays in your browser — we never upload it.
You can drop several images at once — up to five, because each enlargement is a long run. The finished ones come back as a single ZIP.
How can there be pixels where there were none?
A conventional resize guesses the missing pixels from the average of their neighbours. That gives you a bigger picture, not a sharper one: edges blur, and detail cannot come back because nothing knows anything about it. This tool works differently. It runs a neural network trained on hundreds of thousands of image pairs — each one a downscaled copy and its original — so it has learned what a blurred edge, a texture or a grain actually looks like at size.
The model runs in your browser, on WebAssembly. It downloads itself the first time you use it and comes out of the cache after that. The image never reaches a server — not ours, not anyone else’s. That is also why processing is not instant: we cut the picture into small tiles and run the model over them one at a time so memory does not run out. The progress bar is counting exactly those tiles.
What people actually use it for
Old, low-resolution photos. A shot from a 2005 digital camera, or one saved out of a chat, becomes big enough to print without eyes and text melting into mush.
Product shots and logos. The 400-pixel image your supplier sent holds up in a larger storefront frame, and clears the minimum resolution ad platforms ask for.
After a crop. Cut a small detail out of a large picture and the result is unusably small on its own — upscaling gives it its size back.
Where its limits are
Honestly: the model does not perform magic. What it invents is plausible detail, not the truth — an unreadable number plate will not become readable, and a distant face will not become a portrait. It does not fully erase heavy JPEG compression either; from blocky edges it sometimes makes a pattern of its own.
Your device matters too, and not by a little: a few hundred thousand pixels finish in seconds, but a 2-megapixel photo at 4× takes around two minutes even on a desktop, and considerably longer on a weaker phone. That is why we say up front how many tiles a run is, why there is a cancel button, and why we refuse inputs above 16 megapixels: four times that no longer fits in any browser.
Frequently asked questions
Is it really free?
Yes. No daily cap, no sign-up, and no watermark on the result.
Does my image get uploaded anywhere?
No. Processing happens in your browser. It is only uploaded if you click “Save to FileMv" yourself, to get a link to it.
How large an image does it accept?
Up to 16 megapixels, with a long edge of at most 4096 pixels. Above that, upscaling makes little sense anyway — the image compressor is the right tool there.
Why is the first run slow?
Because that is when your browser downloads the AI model. It is a one-off step — the next images skip it.
Should the output be PNG or JPG?
At 4× a PNG easily runs to tens of megabytes. For photos JPG is the saner choice; pick PNG when you still plan to edit the image, or when it is graphics rather than a photograph.