2026-10-02
Choosing between an image CDN and self-hosted delivery is a decision most sites hit once. The intuitive answer is to self-host while traffic is small and move to a CDN once it grows, but traffic volume is not what decides the outcome. What matters is where your visitors are, how many size variants your images need, and whether you want to maintain a transcoding pipeline. This comparison breaks out billing, latency behavior and operational load for both, then gives a decision rule you can apply plus a hybrid setup that avoids the tradeoff.
People tend to read image CDN as servers that cache your images, which is too narrow. A modern image CDN handles at least four jobs.
Of those four, the valuable ones are usually format negotiation and variants rather than distribution. Doing both by hand means maintaining multiple versions of every image and writing correct srcset markup on the front end.
Self-hosting is not free. It moves cost off the invoice and onto a person. The parts you own are listed here.
On a mid-sized site that is a few days of work once, then a recurring check every time a dependency is upgraded.
The largest above-the-fold image drives LCP directly. Differences between the two approaches come down to time to first byte and connection reuse.
If your visitors sit in one country and your images are already compressed, the LCP improvement from a CDN will be smaller than expected. Image byte size tends to matter more than delivery method, and the numbers behind that are in /blog/image-compression-web-performance-guide.
Image CDN pricing usually stacks three meters.
Self-hosting bills as bandwidth plus storage plus servers, plus labor that never shows up on an invoice. The meter that surprises people is requests. Thirty images per page, three variants each, requested once on mobile and once on desktop, and the free tier disappears fast. To estimate, multiply daily pageviews by images per page by average variants to get monthly requests, then check the vendor tiering rather than estimating from monthly bandwidth, which is off by an order of magnitude.
Sites like this buy convenience from a CDN more than performance.
It does not have to be either-or. A practical combination looks like this.
This avoids per-request transform billing without making you run a global network.
If visitors are concentrated in one region and images are already compressed, the gain is limited. Reducing image size usually does more than switching delivery.
No. A CDN handles distribution and transcoding, not an oversized original. A 3MB source image still loads badly on a weak network even behind a CDN.
Pre-generation wins when variant counts are low and access is concentrated. On-demand wins on storage when combinations are many and long-tail access is scattered, but watch request billing.
Usually not a second layer. A full-site CDN already solves proximity. What is missing is format negotiation and size variants, which a self-hosted transcoding pipeline can cover.
Shrink the bytes first, then argue about delivery. image-compressor-saas.shop compresses inside your browser so files never reach a server. Bring the originals down to a sane size first, then decide whether a CDN is worth it, since most sites find that self-hosting is enough once compression is handled. More reads: /blog/image-compression-web-performance-guide, /blog/core-web-vitals-fix-lcp-images, /blog/avif-vs-webp-in-depth.