The Looksmaxing AI score blends an AI model's assessment of the face with a prediction from facial geometry, then turns the features that can change into a potential score and a 90-day plan.
Scoring pipeline
Looksmaxing AI has four components:
Geometry module. It detects facial landmarks on each photograph and calculates measurements from their coordinates with fixed formulas.
AI model. An LLM (large language model) that accepts images and text together. It receives the photographs and the measurements and returns feature scores, an overall score and recommendations.
Geometry model. A statistical model trained to predict the overall score from facial measurements alone.
Plan builder. A rule-based system that turns the feature scores into a 90-day plan from a library of tasks.
01PhotographsFrontal; lateral optional
02Landmarks478 points per photograph
03MeasurementsRatios and angles
04Two modelsAI model and geometry model
05Blended score70% AI model, 30% geometry model
06Potential and planTarget score and 90‑day routine
The sections below follow the pipeline from left to right.
Input photographs
Looksmaxing AI requires a frontal photograph and accepts a lateral photograph as an option.
Why the lateral photograph is optional
All proportional measurements come from the frontal photograph, so the score can be calculated from one image. The lateral photograph adds profile measurements, such as the nasofrontal angle, chin projection and the gonial angle, which make the jawline, chin and nose scores more accurate.
Facial measurements
Looksmaxing AI uses the same geometry module as Rate My Face. We detect 478 facial landmarks and their coordinates with a facial landmark detection model, a neural network that builds a 3D mesh of the face from photographs, and calculate the measurements from these coordinates.
The measurements, reference ranges and formulas are described in How the Rate My Face score works. Looksmaxing AI adds five measurements:
Measurement
Photograph
Reference range
Eye aspect ratio
Frontal
Higher values indicate rounder, larger eyes
Mouth-to-nose ratio
Frontal
About 1.5
Mentolabial angle
Lateral
120°–130°
Nasal dorsum
Lateral
0 means a straight bridge, positive values a hump
Forehead slope
Lateral
Lower values indicate a more upright forehead
Scoring procedure
After the geometry module has calculated the measurements, they go to two models at once. The AI model reads them together with the photographs, and the geometry model reads the measurements alone. The two scores are then blended into one.
1. Model inputs
The AI model receives:
the frontal photograph and, if provided, the lateral photograph
the facial measurements
sex
The geometry model receives the measurements and sex.
Sex selects the reference ranges for the measurements and the reference faces for the overall score. The measurements come in as fixed values, so both models work from the same numbers and the AI model does not estimate proportions or angles on its own.
2. Feature scores
The AI model scores the same 13 features as in Rate My Face, from 0 to 100, and reads them in the same bands. The feature table and the bands are in How the Rate My Face score works.
3. AI model score
The AI model assesses the face as a whole, because overall attractiveness is not the mean of individual features (Rhodes, 2006). It places the face on a scale of reference faces of the same sex, from 3.0 to 9.5, and names the two closest reference points. The scale has 5.0 as the average.
4. Geometry model score
The geometry model predicts the overall score from the measurements and sex. Its prediction rests on the measurements only and is not swayed by the overall impression of the photograph, so it gives a second, independent estimate of the same face.
5. Blended score
The final score combines the two estimates:
S=0.7SAI+0.3Sgeometry
S
final score, rounded to one decimal and kept within 1.0–9.5
SAI
AI model score
Sgeometry
geometry model score
The AI model carries most of the weight because it sees what measurements cannot, such as skin, hair and expression. The geometry model adds stability: on held-out faces, adding it raised the correlation of the score with reference ratings from 0.851 to 0.863.
The gap between the two estimates sets the confidence of the result:
Gap between the estimates
Confidence
Up to 0.8
High
0.9–1.8
Medium
Above 1.8
Low
If the AI model is unavailable, the score comes from the geometry model alone and is marked as low confidence.
Strengths and focus areas
The AI model names 2–4 strengths and 2–4 focus areas, each as a short phrase such as "Sharp jawline" or "Recessed chin". Strengths are the highest-scoring features and focus areas the lowest-scoring ones, so both lists follow the feature scores, and a feature never appears in both.
Potential score
The potential score shows how high the score can rise by improving the features that respond to routine: skin quality, hairline, jawline, cheekbones and masculinity or femininity. Skin, hair and body fat change these features; bone structure does not change with routine, so it is left out.
Pb=S+(10−S)(0.4+b)=207.0−F,0≤b≤0.15
S
current score
F
mean score of the five changeable features, on the 1–10 scale
P
potential score, at least S + 0.6 and at most 9.2
The lower the changeable features, the more of the gap to 10 the potential score covers. A low current score leaves more room, so the target gain shrinks as the score rises.
The lower the score, the larger the target gainPotential minus current score, with the changeable features at 6.5
3.03.0
4.02.6
5.02.1
6.01.7
7.01.3
8.00.9
Recommendations and plan
1. Recommendations
For every feature below 60, the AI model gives the single best action. Actions are sorted into two groups:
Group
Features
Recommendation
Habits
Skin, facial bloating, brows, hair styling, posture, teeth, body fat
Recommendations exclude mewing for bone change, crash diets, testosterone and unregulated supplements. Prescription treatments are marked "consult a licensed professional".
2. 90-day plan
The plan builder assembles the plan from a library of tasks, using the feature scores and the concerns picked before the scan. It follows fixed rules, without the AI model, so the same inputs always give the same plan. Each task carries an evidence level:
Evidence level
Meaning
Evidence-based
Supported by clinical studies
Plausible
Has a known mechanism, limited direct evidence
Anecdotal
Based on reported experience
Tasks with stronger evidence come first; among equal ones, easier tasks come first. The plan is split into phases and introduces tasks gradually over 90 days.
A safety layer checks every task for active ingredients. Prescription treatments such as isotretinoin, finasteride or minoxidil are marked for medical supervision, and tasks with ingredients that should not be combined are flagged as conflicting.
References
Rhodes, G. (2006). The evolutionary psychology of facial beauty. Annual Review of Psychology, 57, 199–226
Hönekopp, J. (2006). Once more: is beauty in the eye of the beholder? Journal of Experimental Psychology: Human Perception and Performance, 32(2), 199–209
Was this helpful?
Thanks for your feedback
All methodology topics
Chats
We use cookies
Some cookies keep the site working, and those are always on. The rest, analytics and marketing, help us understand what’s useful and improve things, but only if you’re okay with it. Privacy Policy