Enlarge JPG and PNG images up to 4× using a neural super-resolution model that runs on your own GPU. It reconstructs edges and texture instead of simply stretching pixels.
Your images are processed on your device and are never uploaded.
This tool is compute-intensive and works best on a desktop browser.
Drop your image here
or drag and drop
Accepts JPG, PNG, WEBP · Up to 512 MB per file
Add a JPG or PNG. Smaller inputs upscale faster.
Choose a scale factor and wait for the model to load.
Download the enlarged result once processing finishes.
Ordinary enlargement averages neighbouring pixels, which produces a bigger blurry image. A super-resolution network has learned what sharp edges, hair, and text look like at higher resolution, so it reconstructs plausible detail rather than smearing what exists.
The image is split into overlapping tiles, each processed separately and blended back together. That keeps peak memory bounded so large images do not crash the tab, at the cost of a longer run.
The model weights — several megabytes — download once and then compile for your GPU. Subsequent runs reuse the cached, compiled model and are substantially faster.
It runs but is slow, and mobile memory limits cap the input size. A desktop browser with WebGPU is dramatically faster. The tool warns you before starting a job your device is likely to struggle with.
No. The model enhances detail that is genuinely present; it cannot recover information that was never captured. Treat heavily upscaled faces as plausible reconstructions, not evidence.