VQS is a way of measuring how good a real estate photograph is, using numbers instead of opinions. This page explains what it measures, how an image gets a score, and what the standard deliberately leaves alone. No prior knowledge assumed.
VQS stands for Visual Quality Score. It is a published standard that defines a fixed list of things to measure in a real estate photograph, exactly how to measure each one, and how to turn those measurements into a single score out of 100.
Two people looking at the same photo will disagree about whether it is good. Run that photo through the same VQS software twice and you get an identical result every time, by requirement. That is the entire point: it turns "this looks a bit dark" into a measurement that can be checked, compared, and audited.
Every image gets a score from 0 to 100 and one of three statuses: Pass, Flag or Fail. A whole listing gets assessed too, because a set of photos can be individually fine and still look inconsistent side by side.
VQS measures whether a property photo is technically well made and believable as a real place, and reports it as a number that anyone can check.
Real estate images are now shot, edited, AI-processed and AI-generated at portal scale, moving through production pipelines that did not exist a few years ago.
The way those images are judged has not changed. It is still someone's eye, applied to a sample, with no written definition of what "good" means. That approach has three problems: reviewers disagree with each other, it cannot keep up with the volume, and nobody can point to a rule when a photo is rejected.
A standard fixes that by writing the rules down first. Once "too dark" has a number attached to it, the argument moves from taste to measurement, and the check can run on every image instead of a handful.
Everything VQS measures is computed from the image data. Nothing depends on knowing where the photo came from, who took it, or what equipment was used.
The first five categories ask whether the image is well made as a photograph. The sixth asks something different and more modern: whether the scene is a believable depiction of a real place.
Overall brightness, blown-out highlights, crushed shadows, the spread of tones from dark to light, and overall contrast.
Colour casts across the image, on surfaces that should be neutral, and through windows, plus whether the balance holds steady from room to room.
Saturation, residual colour casts, vibrance, the plausibility of skin tones and foliage, and traces left behind by heavy colour grading.
Whether vertical lines are vertical and horizontals level, and whether ceiling lines and floor edges sit true rather than leaning.
Marks left by the pipeline rather than the scene: noise, over-sharpening halos, glow around window frames, and badly handled screens.
Whether depicted objects, staging, scale and materials hold together as a real place: fires, pool equipment, furniture, room type and relative size.
Checks that do not apply are marked N/A and removed from the calculation entirely. A bathroom with no windows is not penalised for its window measurements, and a property with no pool is not marked down for pool checks. Not applicable never means zero.
Each individual check produces a score from 0 to 100. Those roll up into a category score, and the categories roll up into the composite: the number out of 100 that people actually quote.
Things measured on a sliding scale, like brightness, are mapped through published breakpoints, so a small deviation costs a little and a large one costs a lot. Yes-or-no checks use a fixed severity scale instead: nothing found scores 100, a minor instance 80, moderate 55, and severe or repeated 20.
The checks in each category are combined into a weighted average, so the measurements that matter more count for more. Anything marked N/A drops out first and the remaining weights are rebalanced, so the category is still scored fairly out of 100.
The six category scores are combined the same way into a single composite out of 100. One failed check does not collapse the composite, but every failure is recorded in the result and stays visible.
Meets the standard. Nothing needs attention.
Usable, but something measurable is off and worth a look.
Falls short of the standard on at least one thing that matters.
The same three bands are used at every level: for a single check, for a category, and for the composite. The exact threshold values and category weights are published by Standard Vision Lab with the canonical runner, so that every certified score can be reproduced against a specific software build and a specific version of the standard.
Buyers do not look at one photo. They scroll a listing, comparing frames directly: room against room, inside against outside, wide shot against detail.
Under that kind of viewing, differences between frames are obvious even when every frame is fine on its own. If brightness, colour or saturation jumps from photo to photo, the listing reads as unreliable, and the shift itself makes people wonder which frame is telling the truth about the property.
So VQS treats inconsistency across a set as a fault in its own right. It measures how much the frames vary from each other across the whole listing, across just the interiors, across just the exteriors, and within each room type. Where one image is the outlier dragging the set around, the report names that image rather than just reporting that the set is uneven.
Set scoring only applies when images are submitted as a set, and a set needs at least two images. The set result is reported alongside the per-image scores and never changes them: an image scores the same whether it is measured alone or as part of a listing.
This is the published sample report: a real six-image listing, measured and reported. Every check is listed with its measured value, its result, and the thresholds it was judged against.
Two things in that panel are worth pausing on. First, the 31 not-applicable results are not failures and not zeros: they are checks that had nothing to measure, such as window measurements in a room with no windows. Second, the two flags are visible rather than hidden. A report that never flags anything is not a quality standard, it is marketing.
The specification is open and free to read. Anyone can see exactly what is measured and how. Certification is separate, and deliberately narrower.
The software that performs the measurements is called a runner. The specification defines what a runner has to do to conform. Standard Vision Lab operates the canonical runner, and only scores from that runner are certification-grade. Scores from any other conforming runner are advisory: useful for internal quality control and perfectly publishable as such, but they cannot be presented as certification and cannot carry the badge.
A single image. The certificate is bound to that exact file using a SHA-256 digest, which is a fingerprint of the file's contents. Alter the image and the certificate no longer covers the altered version, while the original stays certified.
A whole production pipeline. This tests a workflow rather than one picture. Standard Vision Lab chooses the input images, and the applicant cannot preview, filter or cherry-pick them before processing. The images must cover a spread of room types and shooting conditions. If the pipeline materially changes afterwards, such as swapping or retraining a model, the certification no longer covers output made after that change until it is re-tested.
A perfect composite.
Meets the standard.
Measurably short of the standard in at least one area.
Not certifiable. No mark is issued.
Every badge must show the tier and the current score, and must identify the version of the standard the certification was granted under. Badges displayed on the web retrieve the current score when the page loads, so a badge cannot keep advertising a score that is no longer true.
A standard is only credible if it is honest about its edges. These are out of scope by design, not by oversight.
Taste. Composition, styling and artistic merit are not measured. If no objective threshold can be published for something, VQS does not pretend to score it.
Video. Walkthroughs and drone footage are outside this version.
Floor plans and diagrams. Plans and other diagrammatic assets are not still photography and are not scored.
Whether the listing text is true. VQS reads the image, not the description attached to it.
Legal and advertising compliance. VQS is a measurement standard, not a regulator, and makes no ruling in any jurisdiction.
VQS applies to any real estate still image regardless of how it was made. Photographed, human-edited, AI-processed and AI-generated images are measured by the same parameters against the same thresholds. A conforming runner is not permitted to ask how an image was produced or to use that as an input to scoring.
This is a design decision worth stating plainly: VQS is not an AI detector, and scoring well is not a statement that an image is or is not synthetic. It measures the output image only, and takes no position on how any image should be made.
A measurement standard that quietly changes its own thresholds is worthless, because yesterday's score stops meaning anything.
So VQS is versioned. If the parameter set changes, the version changes with it. A score always states the version it was produced under, and remains valid against that version. That is also why the standard is not named after a number of parameters: the parameter set is expected to evolve, and the version number is what carries that, not the brand.
Every certification-grade score records both the version of the standard and the build of the canonical runner that produced it, so any score can be traced back to the exact rules and the exact software that generated it.
The standard is also honest about the limits of agreement. A single runner must be fully deterministic: the same image scored twice gives byte-identical results. Two different implementations are only required to agree within a published tolerance, and near a band boundary they may even land on different sides of it. That is why certification rests on one nominated runner rather than on whichever implementation happened to be used.
The full normative document: every parameter, the measurement conventions, scoring, and the certification framework.
Six images, 108 checks, every measured value and threshold shown. Filter it by status.
Send up to ten images to be measured against the standard by the canonical runner.