About this tool
A small picture is a common problem. A photo sent over WhatsApp arrives squeezed. A logo exists only as a thumbnail. An old scan is a few hundred pixels wide and a form or a print wants more. Stretching it in an editor makes the pixels bigger, and the edges turn into soft stairs.
This tool does the enlargement differently. It runs a small neural network, trained on pairs of low and high resolution pictures, that decides what each patch of the image most likely looked like at twice the size. Straight lines come out straighter, text is crisper than a plain stretch, and flat areas such as sky and skin do not break into blocks.
What it cannot do is know what was not in the picture. If a licence plate is four blurry pixels, the model will draw four plausible ones. It is a good choice for a profile photo, a product picture, a logo or a print that needs to be a little larger. It is the wrong choice for evidence, for a document where every digit matters, or for a face you need to identify.
Everything runs in your browser. The model, about 7 MB, is downloaded from this site once and kept by your browser. The picture is cut into small tiles that are enlarged one at a time, so an ordinary phone can do it, and then joined back together. Nothing is uploaded, and no copy is kept after you close the page.
How to use this tool
- Choose a small picture. A JPG, PNG or WebP up to about half a megapixel, for example 700 × 700 px. Drag it onto the box or browse for it.
- Pick the format to save as. PNG keeps the most detail and any transparency. JPG is smaller but has no transparency. WebP is small and keeps transparency.
- Run it. The first time, a 7 MB model downloads to your device, then your browser remembers it. The picture is then enlarged in small parts, and a counter shows how far along it is.
- Compare and download. You see the new size in pixels and the file size. The result has exactly twice the width and twice the height of the original.
What this tool does not do
- It enlarges to exactly 2 times. There is no other scale.
- The picture can be up to about half a megapixel (700 × 700 px). Larger pictures are refused.
- It reads JPG, PNG and WebP only. HEIC, GIF, BMP, TIFF, SVG and camera raw files are not supported.
- The enlarged detail is drawn by a model, not recovered from the original. Do not use it where a number, a face or a signature has to be exact.
- It is slow on older phones. A picture near the size limit can take several minutes there.
- The first run downloads a 7 MB model.
Common questions
Does it add real detail that was missing?
No. A small model has learned what edges, hair, text and skin usually look like and redraws the picture at twice the size with that in mind. It makes lines cleaner and softer areas less blocky than simple stretching does, but it cannot recover a face or a number that was never captured. Treat it as a better enlargement, not a restoration.
Why is there a size limit on the picture I choose?
The work happens on your own phone or computer, and a neural network is slow on a large photo. Pictures up to about half a megapixel finish in a reasonable time. A picture that is already several thousand pixels wide rarely needs enlarging.
Why did the first run take so long?
The first run has to download the model, about 7 MB, before it can start. After that your browser keeps it, so the next picture starts straight away. If you are on mobile data, use Wi-Fi for the first run.
Is my picture uploaded?
No. The model comes down to your device and the picture is processed there. Nothing about the picture is sent to any server, which is also why it works on a photo you would not want to share.
Can I enlarge it 4 times?
Not in one step. The tool does 2 times only. You can run the result through it again for 4 times, but each pass invents more of the detail, so a second pass looks less natural than the first.
Will it work on a scanned document or text?
It helps with printed text that is slightly small or blurry, because the model has seen a lot of letters. It will not make an unreadable scan readable, and it can change the shape of small characters, so check numbers and names against the original.
My transparent background turned white.
JPG has no transparency, so transparent areas are filled with white. Choose PNG or WebP to keep them.