AI Image Upscaler
AI Image Upscaler enlarges an image 2x or 4x with a super-resolution AI model running in your browser, so the image is never uploaded.
Paste an image (Ctrl+V), drag and drop, or click to upload
Image file · processed on your device, never uploaded
The first image downloads the Swin2SR lightweight AI model (about 8 MB). It is cached afterwards.
At a glance
- Accepts
- PNG, JPG, WebP, GIF and other images your browser decodes
- Outputs
- The image at 2x or 4x the width and height, as PNG, JPG or WebP
- Where it runs
- Entirely in your browser, on your device
- What leaves your device
- Nothing - the file never leaves your device
- Limits
- Output up to 16.7 megapixels (4.2 on phones); first use downloads an 8 MB AI model; adds plausible detail, not real detail - small faces and text can look off; does not fix out-of-focus blur
- Account
- Not required - there is no sign-up
- Price
- Free, with no watermark and no usage cap
- Made by
- Visrut
- Last updated
- 12 October 2026
Make a small image two or four times bigger and keep it sharp. A normal resize just spreads the pixels out, so edges go soft; this tool runs Swin2SR lightweight, an open super-resolution AI model (Apache-2.0 licence, about 8 MB), inside your browser to draw crisp edges and fine texture instead. Your image is never uploaded. Be aware of what it can and can't do: it sharpens and adds plausible detail, but it cannot recover detail the original never had, so small faces and text may look slightly off, and it won't fix a photo that is out of focus or shaken.
How to use it
- Paste (Ctrl+V), drag and drop, or click to upload an image.
- Pick 2x or 4x and press Upscale.
- Watch it fill in tile by tile, then drag the divider to compare, and switch to Actual pixels to look closely.
- Download as PNG, JPG or WebP.
How much better than a normal resize?
We cut 512 x 512 crops from three sharp photos and a photo of a printed receipt, shrank them to a half and a quarter of that size, enlarged them back with this tool and with ordinary bicubic resizing, and measured how close each came to the real crop (PSNR, higher is closer). At 2x the tool came 4 to 6 dB closer than bicubic on the photos and 14 dB closer on the receipt; at 4x, 2 to 4 dB closer on the photos and 5 dB on the receipt. It is a clear improvement, not a miracle: a 4x result is sharper than a stretched image but still softer than a photo taken at that size.
Good uses, and bad ones
- Good: small product photos, old web images, thumbnails, crops of a larger photo, screenshots and graphics that need to print bigger.
- Less good: photos that are already large (there is little to gain), heavily compressed JPGs (it can sharpen the blocky artefacts along with the picture), blurry or shaky photos.
- Not for: identifying people, reading number plates or anything where the exact detail matters. The added detail is a guess.
Frequently asked questions
Is my photo uploaded to a server?
No. The AI model is downloaded to your browser and runs there, on your graphics card with WebGPU or on the CPU otherwise. Your photo never leaves this tab.
Does upscaling add real detail?
No tool can recover detail that was never captured. The model has learned what sharp edges and textures usually look like, so it draws crisp edges and plausible fine texture where a normal resize would only blur. That looks much better, but it is an educated guess: small faces, distant text and fine patterns can come out subtly wrong, so don't use the result as evidence of what was really there.
Can it fix a blurry or out-of-focus photo?
Not really. The model is trained to enlarge small, sharp images. We tested it on full-size photos blurred on purpose and it made them no closer to the sharp originals. It works best on images that are small rather than blurry - thumbnails, old web images, crops, screenshots.
Should I pick 2x or 4x?
2x doubles the width and height (four times the pixels); 4x quadruples them (sixteen times the pixels) by running the model twice. 4x takes about five times as long as 2x. Use 2x unless you need the bigger size.
Why is there a size limit?
The finished image has to fit in a browser canvas, and Safari won't draw one larger than 16.7 megapixels (4096 x 4096). On phones and tablets the limit here is 4.2 megapixels, because a phone runs out of memory or takes many minutes beyond that. Upscaling is meant for small images anyway - a photo straight from a modern camera is already large.
Which AI model does it use?
Swin2SR lightweight x2, a super-resolution model published by Conde et al. in 2022 under the Apache-2.0 licence. The download is about 8 MB, once, and is cached by your browser afterwards.
Why does it work in tiles?
A whole photo through the model at once would need gigabytes of memory. The image is cut into small overlapping tiles, each tile is upscaled on its own, and the overlaps are blended so there are no visible seams. That also keeps memory flat however big the photo is, and lets you watch it fill in.
What formats can I download?
PNG (lossless, largest file), JPG or WebP. PNG keeps transparency; JPG puts transparent areas on white. Some browsers (older Safari) can't save WebP, and the button is hidden there.