Image Resizer
Set new dimensions without the guesswork.
Image upscaler
Make a small or soft photo bigger and sharper, 2x or 4x, with an AI model that runs on your device. Nothing is uploaded.
or drag and drop images here
JPG, PNG, WebP, GIF, AVIF or BMP images · Multiple files supported
Your files stay on your device
- 1Add files
- 2Adjust
- 3Download
How it works
- 1Choose one or more pictures, up to 10 at a time.
- 2Pick 2x or 4x, and the kind of picture: photo, or drawing and illustration. Automatic picks the best model your device can run.
- 3Select Upscale. The first time, the AI model downloads once (5 to 70 MB) and your browser keeps it for next time.
- 4Compare a plain enlargement with the AI result at full size, then download the picture.
What the options do
- Scale
- 2x doubles the width and height, 4x makes each side four times longer. A larger scale shows more invented detail, so use 2x when a photo is already fairly sharp.
- Picture type
- Photo (best quality) is the strongest photo model and works best with a graphics card (WebGPU). Photo (fast) is a much smaller, quicker model that suits older computers and phones. Illustration is for drawings, anime and flat colours. Automatic uses Photo (best) when WebGPU is available, otherwise Photo (fast).
- Format
- PNG keeps every pixel and transparency. JPG and WebP make smaller files. JPG cannot be transparent, so transparent areas turn white.
Privacy
This tool runs in your browser. Your files are never uploaded to a server.
Questions people ask
Is my picture uploaded?
No. The AI model runs inside your browser, on your graphics card when your browser supports WebGPU and on your processor otherwise. Your picture never leaves your device. Only the model file, which holds no personal information, is downloaded once from Hugging Face and kept by your browser.
Does upscaling add real detail?
It adds believable detail, not the true detail that the camera never recorded. Edges, hair, fabric and text get crisper and noise and blocky JPG artefacts are reduced, but fine patterns such as faces far away or small print can be invented. Use it for improving a picture, never as proof of what was in a scene.
Which AI model is used, and can I use the results commercially?
It uses Real-ESRGAN models (RealESRGAN x4plus, realesr-general-x4v3 and the anime 6B model), released under the BSD 3-Clause licence, which allows commercial use. The pictures you create are yours.
Why is there a size limit?
The result must fit in your browser memory. The larger side of the output can be up to about 6000 pixels, which is 40 megapixels in total. A 4x result of a 3000 x 2000 photo is too large, so use 2x or make the photo smaller first.
How long does it take?
With WebGPU, a 1000 x 1000 photo takes a few seconds at 4x. Without it, the small Photo (fast) model takes under a minute for the same photo, and the best model takes much longer. The picture is processed in small overlapping tiles, so the page stays responsive and you will not see seams.
Which type should I choose for screenshots and logos?
Try Photo (fast) first for screenshots, scans and logos with text. Illustration smooths flat colours and can soften small text. If neither looks right, a 2x result is usually the safer choice.