AI face raters mostly measure symmetry, golden ratio proportions and jawline geometry. Of 25 tools audited, 20 name a measurement category, 8 publish a methodology, and 2 disclose how much each measurement counts toward the final score.
What do AI face raters actually measure?
AI face raters measure symmetry, golden ratio proportions, facial thirds and jawline geometry, and 20 of 25 tools (80%) name at least one of those categories on the page. Only 10 name them with real specificity: a defined ratio, a landmark count, or a named angle. The other 10 name a category and never define it. Five name nothing more specific than "facial features."
The named measurements cluster tightly. Six measurement families cover almost every scored dimension in the sample, and the tools that name them are the same handful of tools every time.
- Symmetry, the most frequently named measurement in the audit, scored by TheFaceReport, Lookmax Analyzer, Overchat AI, Pretty Scale Me and AttractivenessTest.net.
- Golden ratio (phi, 1.618), named by PFPMaker, Overchat AI, Golden Ratio Face, Pretty Scale Me and ImageToolsHub.
- Facial thirds, named by TheFaceReport, PFPMaker and ImageToolsHub, with facial fifths named by TheFaceReport alone.
- Canthal tilt, the eye-corner angle, named by PFPMaker, Overchat AI and Looksmaxxing AI.
- Jawline, named by PFPMaker, Overchat AI, Lookmax Analyzer and Looksmaxxing AI, with the gonial angle named by Overchat AI alone.
- Landmark counts, disclosed as 478 by Lookmax Analyzer and Face Shape Detector, 68 by Pretty Scale Me, and "100+" by Looksmaxxing AI.
Which two tools sell the same six dimensions?
Fotor's Pretty Scale and Clipfly's AI Attractiveness Test offer the identical six analysis dimensions in the identical order: Beauty, Handsomeness, Cuteness, Face Shape, Skin Smoothness, Facial Symmetry. Two products, two companies, one word list, same sequence. The likeliest explanation is a shared white label scoring template rather than two independently engineered instruments. Neither page defines any of the six terms.
How many publish a methodology?
8 of 25 tools (32%) publish a methodology explaining how the score is computed, as a dedicated page or a substantial in-page section. The eight are TheFaceReport, PFPMaker, Overchat AI, AttractivenessTest.com, Lookmax Analyzer, Golden Ratio Face, Nanorater and Face Shape Detector. The other 17 present a number and describe the process behind it as an "algorithm" or, in PrettyScale.com's wording, "complex mathematical calculations."
AttractivenessTest.com publishes the strongest methodology page in the sample. Its science page carries 12 numbered references drawn from PLOS ONE, Cognition, CVPR Workshops, arXiv, Behavior Genetics, PLOS Genetics and four other journals. The page states plainly that the score comes from an end-to-end trained neural network rather than named weighted metrics. It also states the average score is designed to land near the same value across Africans, Caucasians and Asians, a bias claim the rest of the category never raises. Declining to decompose a neural network into metric weights it does not use is more honest than inventing them.
Who tells you how much each measurement counts?
2 of 25 tools (8%) publish numeric weights: Lookmax Analyzer and TheFaceReport. 20 of 25 (80%) never disclose how heavily any single measurement counts toward the final score. A further three, Golden Ratio Face, Pretty Scale Me and ImageToolsHub, confirm that a weighted or combined calculation exists and never give the number.
Lookmax Analyzer's weights move with the goal the user selects. Jawline counts for 30% under "masculine," 22% under "feminine" and 28% under "balanced." Symmetry moves between 16% and 20% across the same three goals, while Cheekbones (20%) and Proportions (12%) stay fixed. Every goal column sums to exactly 100%.
The table below holds every published weight set in the category, plus ours. Three rows is the entire population of tools that answer the question "how much does each part count."
| Tool | Published weights | Sums to 100% |
|---|---|---|
| Lookmax Analyzer | Jawline 30 / 22 / 28, Eyes 22 / 26 / 22, Cheekbones 20, Symmetry 16 / 20 / 18, Proportions 12 (masculine / feminine / balanced goal) | Yes, in each goal column |
| TheFaceReport | Symmetry 35, Facial thirds 25, Facial fifths 25, Golden ratio 15 | Yes |
| theattractivenessreport.com (this site) | Symmetry 30, Facial thirds 20, Proportion (golden ratio) 20, Facial fifths 15, Harmony 15 | Yes |
Our own five weights are published on the how the score is calculated page alongside every tolerance constant the formula uses. A weight a reader can see is what turns a score into a claim they can check.
How many claim a percentile they cannot source?
3 of 25 tools (12%) promise a percentile against a population with no sample size, study or source stated: PFPMaker, Looksmaxxing AI (Face IQ) and Pretty Scale Me. Looksmaxxing AI puts it most precisely, claiming its tool "calculates exactly what % of the population you score higher than." Pretty Scale Me tells anyone scoring 81 to 100 that "this is a top-percentile result by mathematical beauty standards." PFPMaker promises "your percentile" in every report and defines no population anywhere on the page.
PFPMaker comes closest to sourcing the claim and still misses. Its page cites Langlois and Roggman 1990 beside a score distribution chart, but that citation covers the bell curve on display, not the placement of any individual score on it.
0 of 25 tools attach a confidence band or a margin of error to a score. The category invented the harder claim, a precise percentile, and skipped the easier and more honest one: how uncertain that number actually is.
What do AI face raters cost?
11 of 25 tools (44%) state that the face rater itself is free. Only 2 of 25 sell a clearly disclosed one-time report with explicit no-subscription language: TheFaceReport at $17.49 and Lookmax Analyzer at $2.99 to $5.99 across three tiers. Nanorater sits between the two models, charging $2.99 to unlock one report and $19 for a 10-photo comparison, on top of 2 free ratings every 24 hours. Most of the free tools sit inside a larger AI photo platform that sells credits or a subscription for something else entirely.
2 of 25 tools are confirmed recurring subscriptions: Looksmaxxing AI at $9.99 to $79.99 per month, and Umax, whose App Store description states that "analysis results require subscription." A third confirmation was not available for most of the rest, since 8 tools show no price at all on the pages audited: PFPMaker, AILabTools, FaceRate.ai, Maxxing.me, PinkMirror, Clipfly, Media.io and Umax. Umax is the one tool whose subscription is confirmed while its price stays unpublished everywhere.
Which pricing page contradicts itself?
Looksmaxxing AI's pricing page promises "one-time payment, lifetime access to this report" in its own meta description, directly above three subscription tiers. Starter, Creator and Professional each bill monthly or yearly in image-generation credits, and each carries a Subscribe button. No one-time option appears anywhere on the page.
What happened to Umax?
Umax, the category's best known subscription face rater, no longer has a working website. umax.app resolves to a domain broker listing ("UMax.app is for sale... $4,990... one-time"), and umaxapp.com is a near-empty placeholder with a contact link and nothing else. The product is an iOS app, "Umax - Become Hot," rated 4.56 stars across 51,416 ratings and published by Improvement Tech LLC. Its business model is disclosed in exactly one public place: the App Store description, which states that "analysis results require subscription. Additional scans require additional purchase." The category's emblem of pay-weekly pricing cannot be compared against its own marketing page, because it does not have one.
What happens to your photo?
5 of 25 tools (20%) process the free scan entirely on-device and claim the photo is never uploaded: TheFaceReport, Lookmax Analyzer, Pretty Scale Me, ImageToolsHub and Face Shape Detector. 16 of 25 (64%) require a server upload for their free or basic score. The remaining 4 could not be confirmed from their own pages.
Only 1 of 25 tools names a fixed deletion window. TheFaceReport states that paid-report photos are deleted within 2 hours. Everyone else uses event language or none at all. Facewow deletes the photo "once you refresh or leave the page," and Golden Ratio Face removes it once the session expires. Overchat AI says images are "automatically deleted after processing" and never says when processing ends. Seven tools make no privacy claim of any kind on the pages audited.
ImageToolsHub makes the sample's strongest on-device claim: "your images don't go to any server for processing. Everything happens on your device," stated in direct contrast to unnamed competitors. The same page masks its per-feature scores behind "Unlock your [X] score" prompts and never says whether unlocking costs money, an email address, or one more click.
Looksmaxxing AI contradicts itself inside one domain. Its tool page says "Your photo never leaves your device." Its own Privacy-First Architecture section says the tool "processes images securely" and that "we don't store your photos for model training." That second wording only makes sense once a server has seen the photo.
Do you need an account to get a score?
13 of 25 tools (52%) require no signup at all to see a first score. Nanorater gates the tool outright with "Sign in to upload," Umax requires an app download, and PFPMaker releases the metric-by-metric report only in exchange for an email address. The remaining 9 could not be confirmed either way, because their login logic loads client side and never reaches a standard page fetch.
The full audit: 25 AI face raters, 8 criteria
The table below is the audit in full: one row per tool, coded from each tool's own public pages. It shows 7 of the 8 coded criteria. The eighth, whether a tool attaches a confidence band to its score, is omitted here because all 25 cells read the same: no. Our own tool sits in the final row under the identical criteria, and it is excluded from every statistic on this page. An audit that grades everyone except its author is an advertisement.
| Tool | Names what it measures | Publishes a method | Discloses weights | Percentile with no source | Account needed | Price | Photo handling |
|---|---|---|---|---|---|---|---|
| TheFaceReport | Yes, 4 named components | Yes, on the tool page | Yes: 35 / 25 / 25 / 15 | No, and universality disclaimed | No for the free scan | $17.49 one time | On-device free scan, paid photo deleted in 2 hours |
| PFPMaker | Yes, 6 defined metrics | Yes, in-page method section | No | Yes: "your percentile," no population defined | Email for the full report | Not disclosed | Unclear: "processed transiently, never stored" |
| AILabTools | Partial, 6 undefined labels | No | No | No | Unconfirmed | Credit based, not disclosed | No claim on the page, upload required |
| Overchat AI | Yes, 4 criteria with named sub-signals | Yes, in-page table and FAQ | No | No | Unconfirmed | Free tool inside a paid platform | "Deleted after processing," no window, upload required |
| Fotor Pretty Scale | Partial, 6 undefined dimensions | No | No | No | No: "you do not need to log in or register" | Free | "Securely handled," no window, upload required |
| Vidnoz | No, "facial features" only | No | No | No | No for this tool | Free | "We don't save your photos on our servers," upload required |
| Airbrush Pretty Scale | No, "features and symmetry" only | No | No | No | No: "no sign-up or payment needed" | Free web tool, paid desktop trial | No claim on the page, upload required |
| PrettyScale.com | No, "complex mathematical calculations" | No | No | No | No | Free | "Not saved or shared," upload or webcam |
| AttractivenessTest.com | Partial, inputs named but not decomposed | Yes, science page with 12 citations | No, and says why: end-to-end neural network | No | No for the main test | Free, app and API not priced | "Processed and then dropped," server side |
| FaceRate.ai | Partial, generic categories | No | No | No | Unconfirmed | Unconfirmed | Unconfirmed, the FAQ answer is not in the page |
| AttractivenessTest.net | Yes, 6 scored dimensions | No | No | No | No: "no sign-ups, no waiting" | Free, no upgrades offered | "Isn't stored, shared, or used," upload required |
| Maxxing.me | Partial, generic pipeline steps | Explainer article citing 2 studies | No | No | Unconfirmed | Unconfirmed | No claim on the page |
| Lookmax Analyzer | Yes, 5 categories and 478 landmarks | Yes, methodology page plus a weights guide | Yes: goal-dependent table, sums to 100% | No, raw scores only | No: "no account or credit card required" | $2.99 to $5.99 one time | On-device free scan, cloud only for the opt-in styling add-on |
| PinkMirror | Partial, "geometric proportions" | No | No | No | Unconfirmed | Unconfirmed | No claim in the static page, upload required |
| Clipfly | Partial, the same 6 dimensions as Fotor | No | No | No | Unconfirmed | Unconfirmed | No claim on the page, upload required |
| Golden Ratio Face | Yes, 5 defined ratios | Yes, in-page plus linked guides | Partial: equal weighting stated, no numbers | No | No: "free, no sign-up" | Free | "Processed in memory," session UUID, server side |
| Looksmaxxing AI (Face IQ) | Partial, generic plus "100+ landmarks" | No | No | Yes: "exactly what % of the population you score higher than" | No for the free basic scan | $9.99 to $79.99 per month | Contradictory: "never leaves your device" beside server-side language |
| Nanorater | Yes, 6-point scoring rubric | Yes, dedicated methodology page | No | No | Yes: "sign in to upload" | $2.99 one report, $19 for a 10-photo comparison | "Not stored on our servers," upload required |
| Pretty Scale Me | Yes, 68 landmarks and 7 named factors | No page, but substantial in-page text | Partial: "weighted combination," no numbers | Yes: "top-percentile result," no sample | No | Free | On-device: "never uploaded to any server" |
| Media.io | No, "features, symmetry, aesthetics" | No | No | No | Unconfirmed | Not disclosed, plan dependent | No claim on the page, upload required |
| ImageToolsHub | Partial, 2 factors and 7 masked categories | No page, but a substantive FAQ | Partial: "combined," no numbers | No | Unconfirmed, scores masked behind unlock prompts | Free, per its own page | On-device: "everything happens on your device" |
| Cutout.pro | Partial, broad undefined categories | No | No | No | Unconfirmed | 5 credits per generation | No claim on the page, upload required |
| Face Shape Detector | Yes, 4 measurements with stated ratio logic | Yes, in-page, names its technology | Not applicable, shape classifier | No | No | Free | On-device: 478 landmarks, "never sent to any server" |
| Facewow | Yes, 4 measured plus 5 vibe labels | No | No | No | No: "no login" | Free | Deleted when you refresh or leave the page |
| Umax | No measurement named anywhere public | No | No | Unconfirmed, visible in-app only | Yes, app download required | Subscription, price not public | No privacy statement on any Umax web page |
| theattractivenessreport.com (this site, not one of the 25) | Yes, 5 named components | Yes, dedicated methodology page | Yes: 30 / 20 / 20 / 15 / 15 | No, percentile and confidence band withheld until studies exist | No for the free scan, email at paid checkout | $19 one time, never a subscription | On-device free scan, paid photo deleted within 24 hours, never used for training |
Read down the weights column: two Yes cells in 25 rows. That column is the category's widest gap and its cheapest one to close, since publishing a weight costs a paragraph and nothing else.
How we ran this audit
We pulled 41 live Google result pages on 19 August 2026 for the search terms in this category: attractiveness test, face rater, am i pretty, rate my face, facial analysis and 36 more. From those pages we took every distinct AI face rating tool that ranked, then read each tool's own pages on 29 August 2026. 25 tools met the inclusion rule: a public page offering a facial attractiveness, beauty or face score to a general visitor.
Coding followed the same five steps for every tool, in the same order.
- Fetch the tool's own page, plus its methodology and pricing pages where those exist as separate URLs.
- Read the page a standard fetch actually receives, since several tools render their terms client side.
- Code eight criteria per tool: measurements named, methodology published, weights disclosed, percentile without a source, confidence band without a source, signup required, price, and photo handling.
- Quote the tool's own wording in the dataset cell whenever a criterion turns on a specific claim.
- Mark a cell uncertain when the page does not answer the question, rather than inferring an answer from the interface.
Uncertain is a finding, not a gap in the work. 9 of 25 tools could not be confirmed on signup, and 4 could not be confirmed on whether a photo leaves the device. In both cases the text that would answer the question loads in the browser rather than in the page. A tool that keeps its terms behind JavaScript has still not disclosed them.
What this audit cannot tell you
- Accuracy: no tool's score was tested against human ratings, so nothing here says which tool measures a face better.
- In-app behavior: no account was created and no purchase was made, so disclosures that live behind a login or inside an app stay outside the sample.
- Time: every cell is one page on one day, and a page can change the day after it was coded.
- Intent: a missing disclosure is coded as missing, never as deception, since the dataset records what a page says rather than what a company knows.
Why our own tool is in the table
We publish our own row because an audit whose author is exempt is marketing. Our five weights are published in full on this site's methodology page: symmetry 30, facial thirds 20, proportion 20, facial fifths 15 and harmony 15, summing to 100. The free scan runs in your browser and the photo never leaves your device. The paid Attractiveness Report is $19 once, never a subscription, and the photo it uses is deleted within 24 hours and never used for training.
Two cells in our row are honest losses. We have no independent validation study, so the percentile and confidence band columns stay empty on purpose rather than carrying a number we cannot defend. And our weights are a production calibration: the formula this instrument actually runs, not a finding about which part of a face matters most to human beings. Every tool in the table could publish that sentence. One does, and it is a competitor: TheFaceReport calls its own weights "this tool's disclosed production calibration, not a scientifically universal formula."
How to cite this audit
Cite this page as: The Attractiveness Report, "The AI Face Rater Audit," 29 August 2026, https://theattractivenessreport.com/research/ai-face-rater-audit. Journalists, researchers and competing tools are free to reproduce the table and the figures with attribution and a link to this page.
- Publisher: The Attractiveness Report, theattractivenessreport.com.
- Sample: 25 AI face rating tools drawn from 41 Google result pages.
- Collection dates: result pages 19 August 2026, tool pages 29 August 2026.
- Dataset: the full CSV at /data/ai-face-rater-audit.csv, 25 rows and 14 columns, including the quoted source text behind every coded cell.
- Corrections: when a tool changes a disclosure, the row is recoded and the change is dated on this page.
The dataset carries more than this page shows: every quote, every uncertain cell with the reason it is uncertain, and the subscription notes for all 25 tools. Anyone who wants to check a number in this audit can recode the row from the same source text we used.