Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A trustworthy website image auditor reports three different things separately: what the browser actually downloaded, what a stated image transformation might save, and what performance tests or real visitors experienced. An estimate can help prioritize image work, but it is not proof of bytes saved or a faster page.
What an image auditor should—and should not—claim
Every result needs an evidence label. “Observed” means the value came from a specific resource response or test run. “Estimated” or “potential” means it came from a model of a change that has not necessarily been made to the site. Field data describes real visits across a reporting population and period.
- Observed resource facts: the tested URL, response bytes, image format, and natural and rendered dimensions. Record the browser, viewport, device pixel ratio (DPR), network and cache conditions, and timestamp where relevant.
- Modeled opportunity: a possible reduction under named transformation settings, such as target format, encoder or quality, and resize dimensions. It is a counterfactual, not a measurement of a changed page.
- Performance results: separately labeled lab-test outcomes or real-user field metrics. Neither follows automatically from an image-level byte estimate.
Use wording such as “How much could this image save under these assumptions?” for a modeled result, “How many bytes did the browser transfer?” for a resource measurement, and “Did visitors experience better performance?” for field or test outcomes. Do not promise a speed increase or a Core Web Vitals pass from estimated image savings alone.
How Lighthouse’s documented image estimate works
Chrome’s image guidance describes a Lighthouse opportunity that collects JPEG and BMP images, recompresses each at compression level 85, then compares the compressed result with the original. The documented audit flags an image when potential savings are at least 4 KiB. These figures and method come from Chrome for Developers’ page dated May 2, 2019; the result is explicitly potential savings, not bytes verified on a modified live page. Chrome for Developers: Efficiently encode images.
#1 Best Overall
The same page says that, as of Lighthouse 13, the audit moved into the “Improve image delivery” insight. Because the guidance page dates to 2019, verify the output for the Lighthouse version you are actually using before describing current report labels or treating this legacy audit as a complete account of current insights.
Chrome’s examples of possible image-delivery approaches include compression, image CDNs, replacing animated GIFs with video, lazy loading, responsive images, correctly dimensioned images, and WebP. They are options to evaluate, not automatic fixes: the right choice depends on the asset, page, quality requirements, and compatibility.
What to put in the report
Keep the measured baseline beside any modeled recommendation so readers can tell what the page delivered from what a proposed transformation might produce.
Recommended Free Tools
| Report item | Evidence label | What to record |
|---|---|---|
| Resource response size | Observed | Bytes reported for the resource response in the measured run, with the implementation’s definition of “transferred” and the relevant browser, cache, and compression conditions. |
| Image dimensions | Observed | Natural and rendered dimensions; include DPR when the sizing rule depends on it. |
| Candidate transformation | Modeled | Source image, target format, encoder or quality setting, resize target, and treatment of transparency or animation. Call the result “estimated” or “potential.” |
| Lab performance | Observed in a test | Results for that run, alongside browser, device emulation, network, cache, and other conditions that can affect it. |
| Real-user performance | Observed field data | Metric, reporting period, population, device grouping, and URL grouping when available. |
If the auditor has not fetched and measured a transformed asset, it should not say “saved X bytes.” If it has fetched the candidate response, identify that response and run as a measurement. A comparison to a hypothetical unoptimized state remains an estimate unless that state was also measured under comparable conditions.
Why a modeled byte reduction is not a speed result
Reducing image bytes can reduce bandwidth demand, but page timing also depends on network conditions, server response, resource scheduling, caching, and the rest of the page. Adding up per-image estimates cannot establish a page-load improvement or a Core Web Vitals outcome.
Google distinguishes field data from individual live tests. The Search Console Core Web Vitals report uses real-world usage data from CrUX, groups similar URLs, and reports LCP, INP, and CLS. PageSpeed Insights and Lighthouse can test individual URLs. These sources answer different questions: field results represent visits within the report’s population and period; a live test describes its particular run. Google’s “Good” thresholds are LCP ≤2.5 seconds, INP ≤200 milliseconds, and CLS ≤0.1. They are page-experience thresholds, not image-savings targets or a promise that image changes will make a site pass. Google Search Console Help: Core Web Vitals report.
Search Console’s URL groups need sufficient data, and Google notes that the report does not list every indexed URL. A PageSpeed Insights result for one URL may therefore differ from the grouped Search Console result. Name the source and scope when reporting or comparing these results.
Free tools Windows power users keep installed
One-click scans. No signup required.
How to turn an opportunity into a verifiable fix
- Capture the baseline. Record the URL, timestamp, browser, viewport and DPR, network and cache conditions, response bytes, format, and image dimensions for the run.
- State one candidate change. Identify the image and specify the proposed format, encoder or quality, and resize target. Note how transparency or animation will be handled.
- Label the output honestly. Until the transformed asset has been fetched and measured, call its possible reduction estimated or potential. If the candidate is measured, identify that response without implying the live page has changed.
- Check visual quality and compatibility. Compare the output at its intended display size and verify transparency, animation, and browser support. If a conversion produces a larger file or unacceptable output, retain the original or choose another approach.
- Deploy and test the actual change. Rerun the page test under recorded conditions and report those results as a new test, not as confirmation that the earlier estimate was a measurement.
- Review field data separately. Inspect real-user metrics over the relevant reporting period; do not treat a single lab run or an image estimate as a substitute.
Google’s older image guidance cautions that a third-party transformation can make an already well-optimized image larger. That page is explicitly outdated and tied to deprecated PageSpeed Insights API v4, so it is useful only as a general reason to compare output rather than assume every conversion helps. Google: Optimize images.
Best Value
Responsive images and “oversized” warnings
A surfaced reader question captures a common ambiguity: why might Lighthouse flag a responsive srcset image as oversized when the browser is choosing a candidate for DPR? A warning is an audit’s opportunity estimate; by itself, it does not establish which candidate a particular browser selected in a particular run. To evaluate the page, inspect the actual resource response and compare the selected image’s intrinsic dimensions with its rendered size and DPR. Treat the warning and the browser’s delivered resource as distinct evidence, rather than assuming either that the audit is wrong or that it independently verified candidate selection.
Build the interface around uncertainty, not a single savings score
A useful auditor makes the evidence visible at a glance. Separate measured response facts from modeled opportunities, put transformation assumptions next to each estimate, and let readers inspect the baseline asset. Avoid rolling modeled byte reductions into a guaranteed site-wide speed figure. When an estimate is uncertain, could damage quality, or could produce a larger or incompatible asset, explain that uncertainty or omit the estimate.
Chrome’s documented approaches—including compression, responsive sizing, lazy loading, image CDNs, WebP, and replacing animated GIFs with video—can guide practical recommendations, but each needs validation against the actual page. The auditor’s job is to identify and prioritize plausible delivery improvements, then make verification possible after implementation.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
