Face Form Finder · output check
Face Analyzer
Read the output before you compare the result.
If any factor is 0, keep the number as a product-specific result rather than converting or averaging it. The ledger below is what that check looks like on one file.
Same image, different numbers
One generated, non-identifiable front-view reference image was run on 28 September 2026. Each number below was printed on that same file, in one sitting, in the two products named across the top of the table. They are not competing scores.
| Recorded field | Lookmax Analyzer | Face Shape Detector |
|---|---|---|
| Headline number printed | 86 / 100 | 93% confidence |
| What that number is about | a face-score preview | a face-shape classification |
| The scale printed with it | a 100-point scale | a confidence percentage |
| Other numbers printed in the same report | 89–91 potential; 82 / 76 / 68 under a heading marked (SAMPLE) | 51% / 46% / 56%; 10% / 97% / 78% / 46% |
| Read again from the same file | not re-read | 92% confidence |
| Run order on the same file | 1 | 2 |
Decision: 86/100 is a face-score preview; 93% is a face-shape classification confidence. A matching denominator does not make distinct outputs interchangeable. Keep them in separate rows unless both tools answer the same requested question.
One report, several scales
Both reports printed more than one number, and the extra numbers are not the headline written again.
The first report put 86 out of 100 at the top. Under a heading its own label marks as a sample, it then printed 82, 76 and 68. The distance between the highest and the lowest of those three is 14 points, inside a report whose headline is one number. The headline alone does not carry that spread.
The second report printed 93% confidence beside readings that run from 10% to 97%. Those readings are not points out of the same hundred as the headline, and the report does not print one scale for both.
A number is readable together with the field it belongs to. Two numbers printed side by side in one report can sit on two different scales, and neither converts into the other.
The same file, read twice
The second report was read from that same file a second time, later in the same sitting. The first read printed 93% confidence. The second read printed 92%.
One point is small, and the size is not the point. The file did not change and the number did. A one-point gap between two reads of one file is not evidence that anything about the face changed, and it is a reason to write the read date beside the number.
The two headline numbers, the other numbers each report printed, the second read of the same file, and the run order, as a CSV. Download face-analyzer-output-ledger.csv
Watch the output check
Prefer YouTube? Watch the face analyzer output check.
Make an output ledger before you read the headline score
Measured: a stated distance, angle, ratio, landmark position, or visible photo feature. Do not substitute a broad score or a style suggestion.
Classified: a named group, such as a face-shape label or feature category. Do not treat it as a measurement unless the tool shows the rule that made the label.
Scored: a product’s summary number or rating. A change in it does not prove that the parts of the report moved with it.
Recommended: an action, style direction, crop, or editing suggestion. It is not evidence that the underlying measurement is correct.
When two results can be compared
First match the requested output. A face-shape label and a symmetry percentage are not competing answers. Then match the photo conditions: camera distance, camera height, face angle, expression, crop, visible hair or glasses, light, and filters. Finally check whether the two products describe comparable methods. A shared number or label is not enough on its own.
Same output, matched capture: repeat the test and compare the stated result. Different output or changed capture: keep both results in separate ledger rows. One result is unnamed or unsupported: retain the boundary rather than converting it into a score.
Do not let an aggregate score answer a narrower question
An overall score can move while leaving you unable to tell which reading moved. Unless a report exposes the changed component, the score change is not an explanation. It also cannot turn a recommendation into a measurement. Choose the narrowest output that actually answers your question and read it on its own line.
Use “not scored” as information
A result that says a photo cannot support a particular reading has drawn a boundary around the evidence available in that capture. Keep that boundary. Do not fill it with a nearby category, a combined score, or a conclusion from another product.
Questions
Why did my face analyzer score change?
First check whether it was the same output on matched photos. A score change by itself does not identify which measurement, category, or recommendation changed.
Can I compare two face analyzer results?
Only when they answer the same kind of question from matched capture conditions and expose compatible methods. Otherwise each result belongs to the product that produced it.
Why did the same photo give two confidence numbers?
Because the read is not fixed. One file read twice in a single sitting printed 93% and then 92%. Write the read date beside the number instead of treating it as a property of the face.
For left-right balance rather than an aggregate output, use the face symmetry test. For a named outline, use the face shape chart.