Image Upscaler
Enlarge a photo or graphic 2× or 4× with AI super-resolution. The model runs entirely in your browser — your image is never uploaded.
Upscaling runs in your browser — your image is never uploaded.
How it works
This tool runs an ESRGAN super-resolution model through UpscalerJS and TensorFlow.js on your device’s graphics card via WebGL. When you click Upscale, the model is fetched once from a CDN and cached. Your image is then processed locally — never uploaded — in small overlapping tiles so it fits in GPU memory, and the network predicts extra detail as it enlarges each tile. The pieces are stitched back into one larger image that you can download as a PNG. The heavy model loads only when you run it, so the rest of the site stays fast.
How to upscale an image
- Drop a JPG, PNG or WebP image, or click to choose one.
- Choose 2× or 4× enlargement.
- Click Upscale and wait for the progress bar to finish.
- Download the enlarged PNG.
Get the best results
AI upscaling works best when it has clean pixels to build on. A sharp, low-noise source — a small logo, an icon, a clear photo — enlarges well, with crisper edges than a plain resize. Heavy JPEG artefacts, blur and noise get magnified too, so a noisy source stays noisy at higher resolution. Start with the smallest scale that meets your need: 2× is quick and often enough, while 4× gives the largest result but takes longer and is best kept to moderate inputs. The model adds plausible detail; it cannot recover information the original never had. To strip an unwanted background from the result, follow up with the background remover.
Why upscale in your browser
Enlarging an image is often something you would rather not upload — a private photo, an unreleased product render, a client’s artwork — and many free upscalers send your file to a server, watermark the output or limit the size. Here the model runs in your own browser tab, so there is no upload, no watermark and no third party touching your image. The trade-off is honest: the first run downloads the AI model (a few megabytes), cached afterwards, and big images or 4× take longer because the network does real work on every tile. Throughout, your image stays on your device. If the enlarged file is heavy, shrink it with the compressor before sharing.
Related tools
Frequently asked questions
Is my image uploaded to upscale it?
No. Upscaling is done by an AI model that runs in your browser using your device’s graphics (WebGL), so the image never leaves your machine. Only the model is downloaded, once, and then cached; the picture itself stays on your device.
Why is the first run slow, and why do big images take time?
The first time you upscale, your browser downloads the AI model — a few megabytes — which is then cached for next time. The upscaling itself runs the image through a neural network tile by tile on your GPU, so large images and the 4× option naturally take longer than small ones at 2×.
How is this different from just resizing?
A plain resize stretches the existing pixels, which looks soft or blocky. This tool uses an ESRGAN super-resolution model that predicts new detail as it enlarges, so edges and textures stay sharper. It cannot invent information that was never captured, but it usually beats a basic upscale.
Is there a size limit?
There is no hard cap, but very large images are slow and can strain your browser’s memory, so the tool scales inputs whose longest side is over 1500 px down to that size before upscaling. For the biggest results, start from a moderate source and use 4×.
Can I use the upscaled image commercially?
Yes. The tool is free and the AI model it runs (an ESRGAN model from UpscalerJS) is released under the permissive MIT licence, which allows commercial use. The enlarged image you download is yours to use.
Image Tools upscales images entirely within your browser. Your image is never uploaded. The first run downloads the open-source AI model (a few MB), which is then cached for offline use.