How to use the AI photo upscaler
- Choose the mode and scale. Photos and general pictures suit Photos and general, while illustrations and simple graphics suit Fast.
- Choose or drop an image. If it is large, the center area is selected first, and you can drag on the image to choose a different part.
- Press Start and wait. When it finishes, compare the result with the plain enlargement and save it as PNG or JPG.
How AI makes a photo sharper
When you simply enlarge a photo, the computer averages nearby colors to fill the new gaps, so blurry outlines and stair-stepped edges just get bigger. AI upscaling is different. A neural network that practiced on many high-resolution photos, blurring them and restoring them, looks at a small photo and estimates what it would have looked like if it were sharp, then draws it in. This tool has a photos-and-general mode and a fast mode, and works by enlarging the image to four times its size.
Calculating the whole image at once would run out of device memory, so the tool splits the photo into small tiles, calculates them one by one and joins them. To avoid visible lines at tile edges, each tile is read together with its neighboring pixels and the seam area is discarded. While it runs you see the tile number and the time left. Choosing 2x shrinks the 4x result in half, so it takes the same time but looks more natural and the file is lighter.
All of this is calculated inside your own device. The photo is not sent to a server, so it never leaves your device. The first time the required files must be downloaded, which takes time depending on your device.
Ideas and limits
You can enlarge an old scanned photo or a picture that became small after being sent through a messenger. It also suits small profile pictures or images that only survive as thumbnails, making them a little larger and crisper. Illustrations and logos with clear outlines usually enlarge cleanly even in Fast mode.
This approach estimates and draws detail that was not there, so it can differ from reality. Small text can turn into similar shapes or smear, and the eyes or skin texture of a face can look different from the original. So do not use it where accuracy matters, such as evidence photos, identity images or medical images.
Processing time depends heavily on your device and the size of the selected area. About 250,000 pixels are processed at a time, and Photos and general takes roughly twice as long as Fast. Enlarging a whole large photo is not supported, and choosing only the part you need is recommended. Selecting just a face or a sign finishes sooner.
A photo that was heavily compressed and shows square block patterns may have those patterns mistaken for texture and enlarged as they are. Results on such photos may be disappointing. On the other hand, moderately small but clean photos often improve noticeably.
When checking the result, turn on the plain enlargement view and flip between the two. You can easily see how much edges, text and hair changed. If you do not like it, change the mode or choose the area differently and try again.
When saving, PNG has no quality loss and is good for keeping, while JPG is much smaller and good for sharing. Images with a transparent background must be saved as PNG to keep the transparency. A result saved at 2x is also smaller than 4x and easier to send through a messenger.
Tips for good results
- Choose a clean original that is not heavily compressed.
- For a large photo, drag to select only the part you want enlarged.
- Try Photos and general first for photos, and Fast for illustrations and logos.
- Choose 2x for screens, and 4x for printing or bigger enlargement.
- Flip between the result and the plain enlargement and save the one you like.
FAQ
Is my photo uploaded to a server?
No. The files needed for upscaling are served as files of this site and all calculation happens inside your browser. The photo is not uploaded anywhere.
Why can it not process a large photo at once?
AI upscaling needs a lot of calculation, and this tool runs in your own browser without a server. So it processes about 250,000 pixels at a time. For a large photo the center area is selected first, and you can drag on the image to choose the part you want enlarged.
How long does it take?
For about 250,000 pixels the photo mode can take one to three minutes depending on your device, and the fast mode takes about half of that. While running it shows tile progress and the time left, and you can cancel. The first time takes longer because the required files have to download.
Does it really make the photo sharper?
Blurry outlines and stair-stepped edges clearly become crisper, but the AI estimates and draws detail that was not there. So fine details of text or faces can differ from reality. Do not use it for evidence photos or identity images.
Should I choose 2x or 4x?
The AI always calculates at 4x, and choosing 2x shrinks that result in half. It takes the same time, but the file is lighter and looks more natural. Choose 2x for photos used only on screen, and 4x for large prints or bigger enlargement.
Notes & limitations
This is a simple free AI upscaling tool and the result can differ from the real photo. Do not use it for evidence, identity or medical images. Photos are processed only in the browser and are not sent to a server. Check the rights before using images that contain other people's works or likenesses.