Image Upscaler
Make a small image bigger without the soft, blocky look of plain enlarging. An open-source super-resolution model runs in your browser and rebuilds detail as it scales — 2× in one pass, 4× in two. Your image never leaves your device.
Private by design: Your image is upscaled by a model running in your browser and is never uploaded. The model is downloaded once from Hugging Face as ordinary site code and cached on your device. See how in-browser processing works.
How to upscale an image
Drop a JPG, PNG, or WebP onto the upload area. Small images are the point — the longest side can be up to 1024 px for 2× and 512 px for 4×.
Choose 2× or 4×. The first run downloads the model, about 8 MB, once; your browser keeps it.
Press Upscale, watch the tile counter, and download the PNG when it finishes.
What upscaling can and can't do
Enlarging an image in an ordinary editor spreads the same pixels over more space, so edges soften and blocks appear. A super-resolution model does something different: it has learned what edges, textures, and text usually look like at higher resolution, and it redraws the enlarged image to match. Lines stay crisp, noise doesn't multiply, and small photos become usable at sizes they weren't before.
It cannot invent information that was never there. A blurry face stays a blurry face at twice the size; text that is illegible stays illegible, just larger and smoother. The best inputs are small but sharp — thumbnails, old web images, icons, scanned line art, screenshots. If the original is out of focus, upscaling will make that more visible, not less.
2× or 4×, and why 4× is two passes
The model here is Swin2SR lightweight, a 2× model released under the Apache 2.0 licence. For 4× the tool runs it twice — the 2× result goes back through the model. We tested a dedicated 4× model and it produced visibly darkened output and failed on repeat runs in the browser, so two honest passes of a model that works beat one pass of a model that doesn't.
Two passes means 4× takes several times longer than 2× and compounds any artefacts, so the input limit is tighter. For most jobs 2× is the better choice: it's fast, it's faithful, and you can always resize the result to the exact dimensions you need.
Why it needs WebGPU, and why there's a size limit
The model works through the image in 256-pixel tiles. On a graphics chip via WebGPU each tile takes around a second, so a 1024-pixel image is a dozen or so tiles and finishes in well under a minute. On the CPU the same tile takes about eighteen seconds and freezes the page while it works — we measured it — so rather than offer that, the tool checks for WebGPU and explains if it's missing. Current Chrome and Edge on a computer have it; recent Safari does too.
The size limit exists for the same reason. A 3000-pixel photo would be hundreds of tiles and many minutes; and a large photo gains little from upscaling anyway. If your image is over the limit, shrink it first or ask whether you need to upscale at all. For a plain enlargement that works in any browser, the Image Resizer does that instantly.
Good to know
Output is always a PNG, so nothing is lost after the model's work. If you need a smaller file, run it through the Image Compressor afterwards.
Faint seams between tiles are possible on very smooth gradients; the tiles overlap to minimise this.
The image is processed on your device. The only download is the model itself, from Hugging Face, and it contains nothing of yours.
Upscaling pairs well with the other tools: remove a background first and upscale the cut-out, or upscale a small logo and then crop it to size.
For what upscaling can and cannot recover, and when 2× beats 4×, read how to upscale an image without losing quality. If a form or printer is asking for "more resolution", how to increase image resolution sorts out which kind it means.
Frequently asked questions
Is my image uploaded to be upscaled?
No. The model runs in your browser on your own graphics chip. The only network traffic is the model file, downloaded once from Hugging Face and cached — you can confirm in your browser's network tab that the image is never sent.
Why does it say my browser can't run it?
Because the upscaler needs WebGPU, and your browser doesn't offer it. Without a graphics chip the model takes about eighteen seconds per small tile and freezes the page, so the tool stops rather than doing that. Current Chrome or Edge on a computer works; so does recent Safari.
Why is there a maximum image size?
The model processes 256-pixel tiles at roughly a second each. A 1024-pixel image is about a dozen tiles; a large photo would be hundreds and take many minutes. The limit keeps the tool honest about what it can do quickly — and large photos rarely benefit from upscaling anyway.
Will upscaling fix a blurry photo?
No. Super-resolution sharpens edges and adds plausible texture, but it cannot recover detail that the camera never captured. A blurry photo becomes a larger blurry photo, with cleaner edges. The best results come from images that are small but in focus.
Does upscaling reduce quality?
The output is saved as a lossless PNG, so the model's result is kept exactly. Whether it looks better than the original depends on the image: sharp graphics, text, and thumbnails usually gain a lot; noisy or out-of-focus photos gain little. Compare the before and after before you download.
Which model is this?
Swin2SR lightweight 2×, an academic super-resolution model (Conde et al., 2022) published under the Apache 2.0 licence with open weights. It runs through the same on-device machinery as the other AI tools on this site.
Guides for this tool
- How to Upscale an Image Without Losing Quality — "Without losing quality" is the phrase everyone searches and no tool can literally promise. Enlarging an image means inventing pixels that weren't there. The real question is whether the invented pixels look right — and modern super-resolution models are genuinely good at that, within limits worth knowing before you start.
- How to Increase Image Resolution — A form rejects your photo for "low resolution". A printer says 300 DPI. A website wants 1200 pixels wide. These sound like the same problem and they are not. Sorting out which one you have is most of the work.
Related tools
Image Upscaler is one of 4 AI tools — remove or blur backgrounds, hide faces, and upscale — models run on your device.