Imagine your product team has commissioned twelve demonstration films from a production partner in China. Each compressed file plays well on its own. On the brand's case-study page, however, a visitor's phone begins requesting multiple videos or players before the visitor has selected a product.

Another compression pass may miss the problem. ONCE recommends commissioning the library's loading behavior alongside its media assets: which user action justifies downloading which resource?

A useful capability with a compatibility qualification

web.dev's Lazy loading video, updated July 2, 2026 and checked October 31, describes native media lazy loading near the viewport. It labels support as limited: browsers without the media loading attribute ignore it. Finding “lazy” in the implementation is therefore insufficient evidence of cross-browser acceptance.

The guide distinguishes video preloading from poster loading. For non-autoplay media, preload="none" addresses video data; it does not universally defer every poster. The Video performance module, updated April 2, 2026, also discusses replacing an immediately loaded third-party embed with a placeholder until interaction.

Those are implementation options and limitations. They do not establish a performance improvement on your site or a universal file-size allowance.

The ONCE three-stage resource budget

This is a proposed acceptance framework, not a browser standard. It separates the cost of presenting choices from the cost of watching a chosen film.

Initial view: make the choices understandable

Identify what must appear when the page opens, before scrolling or clicking. Visible cards need an informative poster, product identification and a reason to watch. Items farther down the page should not automatically compete for the same initial resources.

A video or image serving as the page's main initial visual may need different treatment. Do not apply deferred loading mechanically to that critical asset. A library budget must recognize which content is actually visible on the target device.

Browsing: seeing a card is not choosing playback

Approaching a row may justify its posters or useful preview information. Whether a moving preview is worthwhile depends on its explanatory value and measured device behavior. An automatically looping wall of clips should be a deliberate choice, rather than an accidental consequence of adding cards.

For third-party hosting, the selection card can be separate from the full player. Keep an understandable, operable play entry. A blank rectangle is not an adequate loading strategy, and mouse hover cannot be the only route for mobile visitors.

Playback: load the experience the visitor selected

After activation, show the correct film and a clear loading state. Provide a usable alternative or failure message when playback cannot start. On switching clips, check whether the previous item continues playing or making unexpected background requests.

A poster, preview and full film serve different purposes. A shorter preview must still avoid misleading product identification or removing conditions essential to the claim.

Accept the page, not just the smaller files

Ask the website team to record requests and transferred bytes in three states: open without interaction, scroll to an agreed case, and activate one film. Use agreed devices, browsers, network conditions, cold cache, dwell time and repeatable actions. Retain resource addresses and the page version.

One comparison model is B = initial media transfer + additional transfer during scrolling + additional transfer during playback. Count each actual transfer once and document cache hits and repeat requests separately. This is a project measurement definition, not a full traffic bill or a direct estimate of CDN charges.

  • Many lower-page assets arrive immediately: inspect loading policy before requesting another loss of image quality.
  • Browsing works but playback waits: investigate initialization, media files and delivery, then consider the trade-off of preparing selected resources earlier.
  • Numbers improve but cases are hard to identify: revise the posters and categorization.
  • A browser ignores the attribute: agree and test a suitable fallback rather than declaring universal support.

Include keyboard operation and failure states. No live website change, performance measurement or conversion experiment has been conducted for this article.

Buy a maintainable module

The production partner should supply approved films, listing posters, any required previews, and the mapping of product and language versions. The website team owns loading behavior, player integration and compatibility verification. Operations owns classification and publishing versions. A shared acceptance record connects those responsibilities.

Scope media adaptation, integration and testing, and continuing maintenance as visible work. Agree measurable targets after obtaining a baseline, instead of promising a universal compression percentage or speed result. The thirteenth case should fit an approved module rather than trigger a fresh debate about the whole page.

For ONCE's continuing content services, the reusable asset is the relationship between content purpose, media version, placement, trigger and acceptance record. It lets brands expand a library while keeping the work selectable and usable. ONCE's video production guidance can help turn those requirements into the next supplier brief.