Should you build or buy Image & Media Optimization?
Image and media optimization services resize, compress, and convert images and video to the smallest correct format for each visitor's device, usually through a URL-parameter API sitting in front of a CDN. Engineering teams at content sites, marketplaces, and e-commerce platforms buy or self-host this to cut page load time and bandwidth cost across millions of daily requests.
Copy reviewed 2026-10-04 · Research revision 2026-09-26
Sources: imgproxy.net; imgproxy.net; imgproxy.net; github.com; imgproxy.net
Do we need Cloudinary or imgix to serve optimized images? Not for the transformation and delivery core. Substack, Supabase and Photobucket run self-hosted imgproxy in production, and Thumbor, the open-source image service Globo.com built, does the same job. The work is undifferentiated, and the published case studies put self-hosting well below platform pricing. Buy only if you actually use the DAM, video pipeline or AI editing; if you mostly resize and convert, the platform bill is optional.
Build it, buy it, or bridge?
When building makes sense
Self-hosting the transform core is a documented, mainstream choice once your image volume makes the vendor bill worth optimizing. Photobucket moved off a managed edge-compute optimization service to imgproxy and reported savings of more than 60%, plus a further 20% after moving those workloads to ARM instances. Supabase runs imgproxy on autoscaling AWS ECS at roughly two million images a day. Substack moved off imgix to imgproxy on AWS at about five million images a day and switched its traffic over in about two weeks. Thumbor, the open-source image service Globo.com created, handles the same on-demand resize-and-crop job and lists Wikipedia and Vox Media among its users. These are real production deployments from named engineering teams, not hypotheticals. The work itself, resize, crop, format-negotiate, and cache-deliver, is identical at every company and carries no strategic weight, so there's no reason to avoid self-hosting on competitive grounds. What stays with the vendor is managed video transcoding and an editor-facing digital asset library, which imgproxy and Sharp don't replicate on their own. Smart cropping isn't a vendor-only feature: open-source imgproxy ships simple smart cropping, and the Pro version adds object detection.
When buying makes sense
Buying still makes sense when what you actually use is the bundle around the transforms, not the transforms themselves. Cloudinary, imgix, and ImageKit sell resizing and format conversion alongside digital asset management, adaptive-bitrate video, and AI-assisted cropping or tagging, and teams that lean on those features get real value from the credit-based pricing. The trouble is that many buyers pay for the whole bundle while using only the resize-and-convert core, which is exactly the piece Substack replaced with self-hosted imgproxy when it moved off imgix. Before renewing, check which features your team actually touches: if it's mostly resizing and format conversion, the platform bill is optional, and if it's genuinely the video and DAM workflow, the vendor is earning its fee.
The desk read
Build-versus-buy analysis for Image & Media Optimization is being written. In the meantime, the framework that drives every B4 call is on the B4 Index page.
Frequently asked
Is self-hosting image optimization actually cheaper, or is that just a vendor's marketing claim?
The strongest published figure comes from Photobucket, which reported savings of more than 60% after replacing a managed edge-compute optimization service with imgproxy, then a further 20% after moving those workloads to ARM instances. It's a vendor case study, so treat the exact percentage as directional rather than audited, but Photobucket is a named engineering team running the deployment in production, not a hypothetical benchmark.
What's the actual difference between imgproxy and a service like Cloudinary?
imgproxy is a self-hosted, open-source engine that does resize, crop, and format conversion behind your own CDN. Cloudinary is a managed platform that bundles the same transform capability with digital asset management, video transcoding, and AI-assisted editing, priced through a shared credit pool.
If we self-host, do we lose access to modern formats like AVIF and WebP?
No. imgproxy outputs WebP, AVIF and JPEG XL, and Supabase says it uses imgproxy mainly for resizing and WebP optimization. Format support isn't the gap; managed video and asset-library workflow are.
Does self-hosting mean we give up a CDN?
No, self-hosting the transform layer still sits behind a CDN, whether that's your existing one or Cloudflare or CloudFront specifically. What changes is who runs the resize-and-convert step: your own imgproxy or Sharp deployment instead of a managed vendor's endpoint.