Blog / Technical SEO /
Optimising images for INP and LCP together
You compress every image on the site, run the speed test again, and the button still lags. Annoying. It happens because image optimisation is usually pitched as a file-size job, and that only covers part of the picture.
File size matters a lot for LCP, because a huge hero image delays the main content. INP is a different beast: it measures how responsive the page feels after someone interacts. Images only affect it indirectly, through heavy galleries and scripts. Converting a JPEG won’t fix a slow button. For Core Web Vitals in April 2024, that means LCP, INP and CLS, each a clue about a different part of the experience. Test the page. Don’t apply one generic “image SEO” fix.
Start with the LCP image
The Largest Contentful Paint element is often a big image near the top, though it can be text or something else. Use PageSpeed Insights or Chrome DevTools to see which element counts and when it loads. If it’s an image, check its dimensions, format, transfer size, priority and whether the browser finds it early in the HTML.
Serve an image close to its rendered size, with responsive sources so small screens don’t download a desktop-sized file. Compress while keeping visible quality. WebP can cut bytes, but compare the output and make sure delivery works for the browsers your audience uses.
Never lazy-load the main above-the-fold image. It delays the one image the visitor needs first. Images further down the page are fair game.
Keep layout stable
Cumulative Layout Shift records unexpected movement. An image with no reserved space makes text and buttons jump when it loads. Give the browser the width and height, or reserve the space in CSS, so the layout is stable before the image arrives.
Check responsive images and embeds too. A gallery, product image or third-party video can change the layout after loading. Test a few viewports and watch the page as it loads, not only the final screenshot.
If a banner changes height at different breakpoints, define its aspect ratio or dimensions intentionally. Avoid inserting promotional content above existing content without reserving space.
Understand INP separately
Interaction to Next Paint measures responsiveness across interactions like opening a menu, picking a filter or adding to basket. Image files are rarely the direct cause of slow handling. A heavy gallery script, an image slider or image processing on the main thread can be.
Use Chrome DevTools or field diagnostics to find the slow interaction and the code running at that moment. If a gallery blocks the main thread, look for a lighter version, fewer scripts at once or deferred non-essential work. I wouldn’t compress every image and hope INP improves. Find the real bottleneck.
Choose optimisation based on the page
A blog article needs a featured image and a few diagrams. A product page needs several gallery images with zoom. A portfolio treats big visuals as the content itself. The right image budget differs by template.
Make a checklist per template: maximum upload size, preferred formats, thumbnail generation, lazy-loading rules and alt-text guidance. Keep the originals so you can make new versions later, and don’t recompress an already compressed image, because quality drops each time.
Measure before and after
Pick a representative URL and test with the same device and network settings each time. Record the LCP element, image transfer size and request timing. For INP, use the controls that matter. Field data in Search Console and CrUX shows how real users fare, when there’s enough of it.
After changing image delivery, check the page on a phone and a desktop. Is the right crop loading? Is anything blurry? Do the controls respond, and does the layout hold still? A smaller file isn’t a win if shoppers can no longer examine the product.
Protect the image’s actual purpose
Product images need more detail than an article thumbnail. Shoppers lean on close crops to judge texture, colour or a label, so heavy compression can hurt the sale. Keep a high-quality source, offer sensible gallery sizes and make sure zoom doesn’t pull in the full-resolution original before it’s needed.
For editorial images, a decorative file doesn’t need elaborate alt text, but an instructional diagram does. Describe meaningful information that the surrounding text doesn’t already give. Image work should improve access and page experience together, not turn alt text into a string of keywords.
Set an image workflow for each template
Keep the original in a sensible source folder and let the CMS or image service create the sizes each template needs. Check that the generated srcset and sizes values match how the image is really displayed. A browser can’t pick a useful smaller file if the markup describes the wrong layout. After any theme or CDN change, compare the source URL with the delivered file in developer tools. Two systems may both resize or convert the same image, which adds steps and no benefit.
For a repeatable workflow, write down crop, dimensions and quality for article headers, product galleries and thumbnails. Whoever uploads content should know which file is the source and which is a derivative. That avoids a tiny thumbnail ending up in a big slot, and it makes later design changes easier. Recheck representative pages after updates instead of assuming every template shares one pipeline.
FAQ
Does WebP improve INP?
Usually not directly. Smaller files can reduce network work, but INP concerns responsiveness to interactions. Diagnose the slow interaction and its scripts.
Should every image be lazy-loaded?
No. The main image visible immediately should generally be available promptly; lower images may be lazy-loaded to avoid unnecessary work.
Can image optimisation improve LCP?
It can when the LCP element is an image and image delivery is the bottleneck. Check the actual element and request timing first.
Related reading
- Core Web Vitals in plain English: LCP, FID and CLS
- Page speed on a budget: what to fix first
- Lazy loading images: when it helps and when it hurts LCP
Inspect one page’s LCP element and one important interaction today. Fix what you measure, not what you assume. Which of the two is worse on your site?