Reimagining the image pipeline at Squarespace
summary
Squarespace replaces upload-time resizing with on-demand image processing in the origin, converting and resizing images to WEBP (and serving fallbacks) so variants are generated on cache miss roughly once per year per image. A video-thumbnail pilot proves viability, processing uses whips instead of ImageMagick for lower memory and faster CPU, and the rollout shows ~30% fewer CDN bytes, a 20% faster largest contentful paint and about 50% storage reduction across image buckets.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:29Why images matter
- 3:42Old pipeline overview
- 7:16Pipeline gaps and trade-offs
- 9:23Modern formats and tools
- 12:50Pipeline reimagined: WebP and libvips
- 15:41On-demand processing
- 17:24Pilot project steps
- 21:19CDN format serving
- 24:08Gradual rollout
- 26:57Results and improvements
- 28:47Final takeaways
related talks
44mImage Optimization for eCommerce websites
30mYour own 'Images as a Service'
25mDelivering Fast and Beautiful Images and Video
41mDeliver Perfect Images At Any Size
37mImage Formats, Performance and Cognitive Load
17mEfficient & Frugal Image Data Processing Pipelines
33mImpatience is a virtue
36mDynamic Image Optimization with imgproxy