LCP image size
The LCP element: transfer size and natural / rendered pixel size.
At a glance the headline numbers for LCP image size
The LCP element: transfer size and natural / rendered pixel size.
The typical LCP image ships 100.4 KB over the wire. It carries 0.2 MP for 0.1 MP actually shown.
Distribution & median LCP site count and median LCP at each level of LCP image size — bytes
Passing LCP by LCP image size — bytes which level passes the LCP most often
LCP image size — bytes 100.4 KB. p75 290.3 KB. p99 3.6 MB. At the low end (300–738): LCP 2.0s. At the high end (>p98): LCP 1.8s. computed
Distribution & median LCP site count and median LCP at each level of LCP image size — natural mp
Passing LCP by LCP image size — natural mp which level passes the LCP most often
LCP image size — natural mp 0.2 MP. p75 0.7 MP. p99 7.8 MP. At the low end (0): LCP 2.1s. At the high end (>p98): LCP 1.7s. computed
Distribution & median LCP site count and median LCP at each level of LCP image size — rendered mp
Passing LCP by LCP image size — rendered mp which level passes the LCP most often
LCP image size — rendered mp 0.1 MP. p75 0.2 MP. p99 1.0 MP. At the low end (0.002–0.004): LCP 1.9s. At the high end (>p98): LCP 1.6s. computed
Why this matters for the Core Web Vitals, and where to start fixing it
Three numbers describe the LCP image: the bytes over the wire, the pixels in the file, and the pixels actually shown. The gap between the last two is pure waste. An image with far more natural pixels than rendered pixels shipped resolution nobody saw, and the extra bytes stretched the load time.
Right-size first, then compress, then pick the format. Resizing the image to what is displayed saves more than recompressing a wrong-sized original ever will.
How does the weight of the LCP image affect the LCP?
LCP image size correlates with the LCP. Displayed pixels separates passing sites from failing sites the most. Where the displayed pixels is low, 69% of sites pass the LCP. Where it is high, 82% do. The rise is gradual.
Chrome field data from 94,910 sites, representing millions of real page loads. How we measured.