How to upscale an image
- Drop in your images, choose them, or paste one.
- Pick 2× or 4×. The first image takes longer while the AI model downloads, then each picture is enlarged piece by piece.
- Drag the slider and zoom in to compare with the original, then download the PNG, or all of them as a zip.
2× or 4×
Both settings use Swin2SR, an open-source super-resolution model. 2× uses its lightweight version, trained to undo plain downscaling: quick, and good at crisp edges in screenshots, illustrations and clean photos. It keeps whatever is in the picture, JPEG blockiness included, just sharper.
4× uses Swin2SR’s real-world version, trained on photos with blur, noise and JPEG compression. It cleans up those flaws as it enlarges, so it suits small, compressed pictures such as images saved from social media or old phone photos. The price is a smoother look: fine texture such as skin and fabric gets softened.
Upscaling versus resizing
Plain resizing, as in resize image, fills the new pixels by blending their neighbours, so edges turn soft and compression blocks grow with the picture. An AI upscaler has learned what sharp edges and textures look like at higher resolution and draws them in. The difference is clearest on outlines, lettering and hair; on smooth areas such as sky the two look much alike. Zoom into the comparison to judge it on your own picture.
Tips
- Start from the best copy you have. Each round of JPEG compression adds blur the model has to guess around.
- Crop first. Upscaling only the part you need is faster and keeps you inside the size limit. Use crop image.
- Save space afterwards. The result is a lossless PNG. For the web, run it through compress image to get a much smaller JPG or WebP.
- Transparent PNGs keep their transparency. The see-through areas are enlarged smoothly alongside the picture.