1. Inventory the uses before exporting
Choose a representative page and list its images: main visual, cards, portraits, gallery and any diagrams. For each placement, record the intended message, file owner, verified usage rights and relevant languages. Keep a high-quality original; website versions should remain identifiable exports.
Create a brief record for each use: page, placement, expected crop on mobile and desktop, proposed alternative text, delivered variants and person responsible for acceptance. If a photograph appears in several cards, inspect every context. This is a suggested handover method, rather than a rule requiring one format for every website.
2. Separate resolution from composition
Responsive images offer the browser several resources. The HTML specification describes srcset and sizes for resolution variants. It also distinguishes art direction: providing a different composition when the available space changes. A smaller version of the same file and a differently cropped image therefore solve different problems.
Use a fictional product photograph as an acceptance example. On desktop, both the object and its setting are visible; on mobile, the subject must remain recognisable. Approve this crop in the design first. Then provide variants that match the planned placements, instead of exporting every image at the width of the largest screen.
3. Compare formats using real content
The picture element can offer sources according to supported format or display conditions, with img providing the fallback. This allows alternative formats or crops. MDN lists WebP and AVIF among the options; the browsers targeted by your project still need to be checked.
Compare exports with equivalent dimensions and intended use. Examine transferred bytes, then gradients, edges, small details and transparent areas where relevant. Keep the file that meets the agreed visual need. Avoid promising a fixed saving or imposing one universal byte limit: a small portrait and a detailed diagram serve different purposes.
4. Find the main image before deferring loading
LCP measures when the largest visible image or text block is rendered. Identify that element in a page measurement; it is not necessarily the photograph the team calls the hero. Local tests and data from real visits provide different perspectives and should be distinguished.
Lazy loading defers fetching images far from the viewport. web.dev advises against applying it to images likely to be visible on opening, particularly the LCP image. In acceptance testing, start a fresh session, inspect requests and confirm the main visual does not wait for scrolling. Adjust resource priority only where you have established the need.
Reference : web.dev — Optimize Largest Contentful Paint · Reference : web.dev — Browser-level image lazy loading
5. Preserve space and meaning
Providing image dimensions or an appropriate aspect ratio reserves space and helps limit content movement when an image arrives. Check alternative crops in the integrated page too: the reserved space should match the intended display without stretching the subject.
Base alternative text on purpose. WAI distinguishes informative images, functional images and decoration: describe the meaning or action, and use empty alt for decoration. For a complex chart, make its information available elsewhere on the page too. Review alternatives in every language and avoid adding keywords unrelated to the content.
Reference : web.dev — Optimize Cumulative Layout Shift · Reference : W3C WAI — An alt Decision Tree
6. Check resource selection and loading
The image element exposes currentSrc to identify the selected resource. Ask your team to record that URL along with the test viewport width and screen density. Selection does not establish successful loading: also check the request response and visible rendering. Seeing a filename in the delivery folder does not prove the browser used it.
Repeat the visit with an empty cache and then a warm cache. Change the width and open the page directly, rather than using internal navigation alone. Record missing files, cut-off subjects, distracting blur and movement during loading. Compare measurements under equivalent conditions; a single fast run does not demonstrate a lasting improvement.
7. Hand over checks the team can repeat
Include approved files, their usage records and observed results. The following grid is a starting point: adapt devices and viewport sizes to the website’s audience. It is neither a full accessibility audit nor a guarantee of performance scores or search rankings.
For each issue, record the page, conditions and expected result, then repeat the scenario after correction. After release, check the real public URL and served resources. Retain originals, the delivered version and acceptance evidence so the next update starts from a clear baseline.
Image acceptance grid to adapt to your project
| Test | Result to check | Useful evidence |
|---|---|---|
| First mobile visit | Recognisable subject; main visual loads without scrolling | Screenshot, viewport and requests |
| Desktop and high-density screen | Suitable resource and readable useful detail | Selected URL and conditions |
| Slower loading in preview | Reserved space and stable neighbouring content | Loading recording |
| Blocked or unavailable image requests | Meaning or action remains clear with the intended alternative | Observed text and context |
| FR, EN and AR versions | Crops and alternatives suit each page | URL, language and observation |
Frequently asked questions
Should every file be converted to the same format?
Decide by use and expected compatibility. Compare real exports and keep a fallback where your delivery strategy requires it. Renaming a file extension does not convert its contents.
Is a good local result enough for acceptance?
It helps diagnose the tested page. Add visual checks, verification of served resources and real-visit data when available. Do not present an isolated measurement as a promise for every visitor.
Where should we start with images already online?
Choose one page template and its main image. Record a baseline, address an identified issue and check the same journey. Then expand to other uses, rather than replacing the entire library without an inventory.
Prepare the material for your team
Bring a representative page, available originals and priority devices. Agree which variants to prepare, who owns each decision and what evidence is needed for acceptance.
Primary sources and limits
WHATWG, MDN, web.dev and W3C WAI documents checked on 30 October 2026. Handover records, examples and acceptance scenarios are editorial recommendations. No Dev Ring performance result or numerical saving is claimed.
