A Core Web Vitals tip: never lazy-load the image that decides your LCP (Largest Contentful Paint). Give it fetchpriority="high", leave loading out, and add a preload only when the HTML can't show the image.

The 2025 Web Almanac still finds 16–17% of pages lazy-loading their LCP image. Only 17% of mobile pages with an image LCP use fetchpriority="high".

What each attribute actually does

  • loading="lazy": the download waits for layout. On a hero, that wait is pure delay.
  • loading="eager": the default. Writing it changes nothing.
  • fetchpriority="high": the request starts at High instead of Low. Google Flights went from 2.6 s to 1.9 s LCP with it (web.dev).
  • lazy + high: still waits for layout. The delay stays.

When you need a preload

Only when the preload scanner can't see the image:

  • a CSS background-image hero
  • an image a script inserts
  • a JS-built slider

Put fetchpriority="high" on the preload too. Match imagesrcset and imagesizes to the element exactly, or the browser downloads twice.

Then cut the bytes

Same hero photo, 1920 px wide, our engine at Auto quality:

  • JPEG: 140.3 KB
  • WebP: 85.5 KB
  • AVIF: 69.3 KB

A phone that shows the hero at 400 CSS px on a 2x screen needs 800 px. That file is 35.2 KB. srcset + a real sizes value gets it there. Note that sizes="auto" only works with lazy images, so your LCP image needs a real value.

Which fix shortens which part of LCP

The markup

<img src="/img/hero-1280.jpg" srcset="/img/hero-800.jpg 800w, /img/hero-1280.jpg 1280w, /img/hero-1920.jpg 1920w" sizes="100vw" width="1920" height="1076" alt="..." fetchpriority="high"> 

Wrap it in <picture> with an AVIF <source> first, then WebP.

What does your LCP breakdown look like? I'm curious which sub-part hurts most on real sites: load delay or load duration.

Full write-up with every byte count: https://blinkimg.com/blog/image-optimization?utm_source=devto&utm_medium=social&utm_campaign=lcp-checklist