Start with the actual layout slot
Measure how widely the subject appears rather than starting with the original photograph’s maximum size. A narrow card and a full-width section may need different deliveries. Screen pixel density also affects the image data useful for sharpness.
For an illustrative card around 300 CSS pixels wide, a display with twice the pixel density may benefit from a candidate larger than 300 image pixels. This is planning input, not a guarantee that the browser always chooses a particular file. Test the actual page and request.
Keep candidates consistent
Resolution variants should normally show the same subject. Cropping that changes the message needs a separate editorial decision. A smaller CMS screenshot should not accidentally show another panel or language.
HTML can describe candidates with srcset and the intended layout size with sizes. Those values need to agree with the layout and real dimensions. Labeling a small image as a large candidate does not make it sharper.
| Check | Evidence |
|---|---|
| Displayed width | Measure the intended layout slot |
| Candidates | Verify real dimensions |
| Subject | Compare panel, crop and language |
| Requested file | Inspect the browser’s network request |
Run a comparable test
Open a safe preview at narrow mobile, tablet and desktop widths. Record displayed size, requested resource and whether the important information remains understandable. Do not confuse a cached large image with a new request when comparing cases.
Try the enlargement link where one exists. A small preview may be appropriate, but readers need access to the correct detailed subject. A dialog showing another language means delivery is incomplete even when the thumbnail is correct. Record both resources.
Assess quality and delivery cost together
Smaller files are not an end in themselves if product labels become unreadable or diagrams lose information. Compare comprehension and delivery in context. Dimensions can help reserve layout space; loading priority depends on the page’s visible content. Keep resource decisions and test cases in the media register. A successful local check is not measured real-user performance. Revisit the evidence when the layout or original subject changes rather than assuming the old candidates remain appropriate.