Upscale Image

Small images look soft and blocky when enlarged with ordinary scaling. Super-resolution models are trained to add plausible detail as they enlarge, making edges crisp and textures convincing. This tool offers both an on-device AI model and classic high-quality scaling, so you can pick what suits each image.

The AI method

The AI option uses Real-ESRGAN, a widely used open-source super-resolution model, in its compact general-purpose version of about 5 MB. It runs on your device: the image is processed in overlapping tiles, which keeps memory use low and lets the page show progress, and the tile borders are trimmed so no seams appear. The model is downloaded once from this site and cached; the image itself is never uploaded.

The classic method

Lanczos resampling is the high-quality interpolation used by professional editors. It enlarges smoothly without inventing detail. It is instant, works for any image size, and is the honest choice when accuracy matters more than crispness, for example for scientific images or documents.

Choosing the factor

Enlarge by 2, 3 or 4 times in each direction. A 500 × 500 image becomes 1000, 1500 or 2000 pixels wide. Larger factors make more visible the difference between methods. For printing, calculate the needed pixels first: a 6-inch-wide print at 300 pixels per inch needs 1800 pixels.

The one-megapixel limit

AI upscaling is computationally heavy. To keep processing times reasonable on ordinary computers and phones, AI mode is available for images up to about one megapixel. Larger images automatically use the classic method, with a notice explaining why. Cropping to the important part, such as a face in a group photo, often brings an image under the limit.

Comparing results

A split slider shows a simple enlargement of the original on the left and the upscaled result on the right, both at the new size. Drag it across details such as eyes, text and edges. Try both methods on the same image; for some pictures the classic result looks more natural.

What works well

Small web images, old low-resolution digital photos, screenshots, logos and illustrations with clear lines, product thumbnails and game art all benefit. Faces become smoother and sharper, foliage and fabric gain texture.

What to watch for

The AI can over-smooth skin, turn noise into odd textures, and alter small text, which it does not read. Check names, numbers and faces before using the result for anything important. Heavily compressed JPEGs with blocky artifacts may enlarge the artifacts too; a light Denoise Image pass first can help.

Transparency

Transparent PNGs keep their transparency; the model enhances the colors while the alpha channel is enlarged from the source.

Enlarging for print

Photo labs print sharply at around 300 pixels per inch, and large posters viewed from a distance can go as low as 150. Upscale a small image by the factor that reaches the needed size, then check details at full zoom before ordering. If 4× is still not enough, upscaling the result again with the classic method adds size without new detail.

Old family photos

Scans of small prints often benefit greatly. Scan at the highest resolution your scanner offers first, since real scanned detail always beats predicted detail, then upscale only if the scan is still too small.

Related tools

For pixel art, use Pixel Art Converter, which enlarges with hard edges. To make images smaller instead, the Image Resizer offers exact dimensions, and Sharpen Image adds crispness without enlarging.