A traditional Content Delivery Network (CDN) improves performance by reducing the geographical distance between website visitors and website servers. By staging content closer to the visitor, the CDN can deliver faster, reduce latency, and avoid long trips back to a single origin server that might be on the other side of the globe. In contrast, ShimmerCat provides a global image CDN. It builds upon a traditional CDN’s geographic proximity approach, but it also adds several innovative improvements. For example, each ShimmerCat edge server always serves the smallest image supported by the browser among several formats (AVIF, WEBP, JPEG, JPEG XL) The second improvement is adding image optimization to the CDN process. When an image is requested for the first time, ShimmerCat will pull the original image from the website’s storage (also called the “origin”). There are two stages: Something we call the online adjuster, that should kick in already on first access, and The optimizer proper. The goal of the online adjuster (1) is to resize the image to a given pixel size, so that we never return a full-resolution original when there is no need (since we fetch the original image to perform the optimization). However, this stage doesn’t do complicated optimization logic, since it needs to run in the edge. As soon as an image is detected and returned the first time, it is submitted to our online optimization service (2). That service measures the degree of JPEG damage already present in your image and uses it to select a quality threshold, and then tries several compression settings and measures the differences to the original image, until we find a format which is the smallest above a certain quality setting. These stages generate compressed images as JPEG (with Mozjpeg), AVIF, WebP and JPEG-XL. Because of these stages, it is normal that you will see images of different sizes and types during the first few requests.