Recipes

Complete visibility-aware React patterns built with react-intersection-observer.

Each recipe states the problem, shows a complete component, and calls out the part that is easiest to get wrong. For the API details behind the examples, see Core APIs and Configuration.

Lazy image loading

Reach for the observer when you need a specific preload margin, a custom placeholder, or a loading transition on the component.

import { useInView } from "react-intersection-observer";

type LazyImageProps = {
  src: string;
  alt: string;
  width: number;
  height: number;
};

export function LazyImage({ src, alt, width, height }: LazyImageProps) {
  const { ref, inView } = useInView({
    rootMargin: "200px 0px",
    triggerOnce: true,
  });

  return (
    <div ref={ref} style={{ aspectRatio: `${width} / ${height}` }}>
      {inView ? (
        <img src={src} alt={alt} width={width} height={height} />
      ) : (
        <div aria-hidden="true" className="image-placeholder" />
      )}
    </div>
  );
}

Reserve the image’s space with dimensions or an aspect ratio. Without it the image shifts the page when it loads, and the observer keeps chasing a moving target.

Scroll-triggered animation

Keep the animation in CSS and use triggerOnce when the reveal should happen only once:

import { useEffect, useState } from "react";
import { useInView } from "react-intersection-observer";

export function Reveal({ children }: { children: React.ReactNode }) {
  const [enhanced, setEnhanced] = useState(false);
  const { ref, inView } = useInView({
    fallbackInView: true,
    threshold: 0.2,
    triggerOnce: true,
  });

  useEffect(() => {
    setEnhanced("IntersectionObserver" in window);
  }, []);

  return (
    <div
      ref={ref}
      className={enhanced && !inView ? "reveal reveal-pending" : "reveal"}
    >
      {children}
    </div>
  );
}

.reveal {
  opacity: 1;
  transform: none;
  transition: opacity 250ms ease, transform 250ms ease;
}

.reveal-pending {
  opacity: 0;
  transform: translateY(1rem);
}

@media (prefers-reduced-motion: reduce) {
  .reveal {
    opacity: 1;
    transform: none;
    transition: none;
  }
}

The content stays visible until the observer takes over, and fallbackInView: true keeps it visible when observation never arrives. Never make the animation the only way to find important content. Reduced-motion users see it right away.

Track an impression

Use useOnInView when the result is an effect and a state update is unnecessary:

import { useOnInView } from "react-intersection-observer";

type ImpressionProps = {
  id: string;
  record: (id: string) => void;
};

export function Impression({ id, record }: ImpressionProps) {
  const ref = useOnInView(
    (inView) => {
      if (inView) record(id);
    },
    { threshold: 0.5, triggerOnce: true },
  );

  return <article ref={ref}>Tracked content</article>;
}

The package ignores the initial false notification, so this callback records the first accepted enter transition. Keep the identifier stable, and check that triggerOnce matches what your analytics counts as an impression.

An intersection does not prove anyone actually looked at the content. Add a minimum-duration rule if your definition requires time in view, and look at Observer v2 if covered or filtered content matters to you.

Infinite scrolling

Observe a sentinel at the end of the list, guard requests while one is in flight, and keep a real button for keyboard and assistive-technology users:

import { useState, useTransition, type Key, type ReactNode } from "react";
import { useInView } from "react-intersection-observer";

type Page<T> = { items: T[]; hasNextPage: boolean };

export function InfiniteList<T>({
  getKey,
  loadPage,
  renderItem,
}: {
  getKey: (item: T) => Key;
  loadPage: (skip: number) => Promise<Page<T>>;
  renderItem: (item: T) => ReactNode;
}) {
  const [items, setItems] = useState<T[]>([]);
  const [hasNextPage, setHasNextPage] = useState(true);
  const [failed, setFailed] = useState(false);
  const [isPending, startTransition] = useTransition();

  function loadMore() {
    if (isPending || !hasNextPage) return;

    setFailed(false);
    startTransition(async () => {
      try {
        const page = await loadPage(items.length);
        setItems((current) => [...current, ...page.items]);
        setHasNextPage(page.hasNextPage);
      } catch {
        setFailed(true);
      }
    });
  }

  const { ref } = useInView({
    rootMargin: "400px 0px",
    skip: isPending || failed || !hasNextPage,
    onChange: (inView) => {
      if (inView) loadMore();
    },
  });

  return (
    <>
      <ul>
        {items.map((item) => (
          <li key={getKey(item)}>{renderItem(item)}</li>
        ))}
      </ul>

      <p aria-live="polite">
        {isPending ? "Loading more items…" : null}
        {failed ? "Could not load more items." : null}
      </p>

      {hasNextPage ? (
        <>
          <div ref={ref} aria-hidden="true" />
          <button disabled={isPending} onClick={loadMore}>
            {failed ? "Try again" : "Load more"}
          </button>
        </>
      ) : null}
    </>
  );
}

There is no effect here, and that is the point. An empty list puts the sentinel inside the viewport, so the observer asks for the first page through the same path as every page after it. An effect for the first page would duplicate the fetch, the error handling, and the pending flag, then race the observer for the same request.

useTransition owns that pending flag. React raises isPending when startTransition runs and lowers it when the awaited work settles, so there is no setLoading(true) to pair with a setLoading(false) on every exit path. startTransition also awaits the async function you hand it, so nothing floats. Marking the append as a transition is worth it on its own: re-rendering a list that has grown to hundreds of rows no longer blocks a click or a keystroke.

A hand-rolled loading boolean is also less reliable here. When loadPage resolves from a warm cache without ever yielding, React can batch the update that sets the flag together with the one that clears it. The observer then never sees skip change, and the list stops after one page. isPending renders either way.

skip does two jobs. It keeps a second request from starting while one is in flight, and because the hook drops and recreates the observer when skip flips back, it re-arms the sentinel. A page too short to push the sentinel out of view keeps loading until the viewport fills.

rootMargin starts the request before the sentinel reaches the viewport. Tune the 400px value to your network and item size. This example sends items.length as the next page’s skip; use whatever pagination your API expects.

Keep the button. It retries after an error, it is the only route for keyboard and assistive-technology users, and it is the guaranteed way forward if the observer never fires again. Scrolling should never be the only way to fetch content.