The six scoring dimensions
When you upload a photo, a vision model reads six visible dimensions — things that can be observed in the image. Each dimension is rated on a 0–10 scale. These are observations about the photo, not a verdict on the person.
| # | Dimension | What it measures | Why it matters |
|---|---|---|---|
| 1 | Lighting | How light falls on the face. | Poor lighting flattens structure and hides detail. |
| 2 | Angle | Camera angle relative to the face. | A low or high angle distorts proportions. |
| 3 | Expression | How relaxed or forced the expression reads. | A tight expression can make the read feel less natural. |
| 4 | Image quality | Sharpness, blur, compression, and exposure. | A blurry or artifact-heavy image hides visible information. |
| 5 | Face visibility | How much of the face is unobstructed. | Hair, glasses, or shadows can block key features. |
| 6 | Harmony impression | How the features read together in this photo. | This is the core PSL dimension — a conservative, non-extreme read of overall facial harmony. |
The first five dimensions measure photo condition quality — whether the image gives the model enough to work with. The sixth dimension, harmony impression, is the only one that reads facial structure directly. Together they produce a score that reflects both the photo and the face.
The weighted formula
The vision model describes the dimensions. The backend computes the score. The two jobs are kept separate on purpose.
| Dimension | Weight | Role |
|---|---|---|
| Harmony impression | 0.45 | Core PSL dimension — the primary signal. |
| Angle | 0.15 | Camera angle shifts how proportions read. |
| Lighting | 0.10 | Photo readability baseline. |
| Expression | 0.10 | Naturalness of the face in frame. |
| Image quality | 0.10 | Sharpness and technical clarity. |
| Face visibility | 0.10 | Completeness of the facial read. |
Why harmony impression carries the most weight. In PSL-style analysis, overall facial harmony is the dimension that most directly reflects what the scale is designed to measure. The other five dimensions are gatekeepers — they measure whether the photo is readable, not whether the face is harmonious. A photo with poor lighting or a bad angle will produce a lower harmony impression through no fault of the face itself. That is why multiple photos matter (see below).
Why angle is second. Camera angle is the single largest photo-condition variable that shifts proportion perception. A low angle elongates the lower face; a high angle compresses it. Giving angle a higher weight than the other photo-condition dimensions reflects its outsized impact on how facial structure reads.
The formula in full:
raw_score = lighting × 0.10
+ angle × 0.15
+ expression × 0.10
+ image_quality × 0.10
+ face_visibility × 0.10
+ harmony_impression × 0.45
psl_score = clamp(raw_score × 0.8, 1.0, 8.0)
If a dimension cannot be read (e.g., the model cannot assess lighting), it defaults to 5 — the midpoint. The formula never invents data.
From raw score to the /8 scale
The raw weighted score ranges from 0 to 10. PSL conventions use a 1-to-8 scale, so the raw score is scaled by 0.8 and clamped to the 1.0–8.0 range:
- A perfect 10 raw score maps to 8.0 — the ceiling of the PSL scale.
- A raw score of 1.25 or below maps to 1.0 — the floor.
- Most readings fall between 3.0 and 7.0, and the final score is rounded to one decimal place.
The 1.0 floor and 8.0 ceiling reflect a deliberate choice: no one is a zero, and no one is a perfect score. The scale is a tool for comparing your own photos, not for ranking people.
Why the model doesn't invent your score
Many AI rating tools are black boxes. A photo goes in, a number comes out, and you have no way to know how the number was produced. If the same photo produces a different score on a second run, there is no explanation.
PSL Range works differently:
- The model reads dimensions. It describes what it sees — lighting quality, angle, expression, and so on. It never outputs a final score.
- The backend applies a fixed formula. The same dimensions always produce the same score. There is no randomness, no model "mood," and no hidden adjustment.
- You can see every dimension. The result dossier shows each dimension score alongside the final estimate, so you know exactly what drove the number.
This split — model describes, backend computes — is the foundation of the transparent scoring approach. It is what makes multi-photo comparison possible and what makes the stable range trustworthy.
Why a single photo is not enough
The fixed formula keeps the math consistent, but it cannot fix a noisy photo. Lighting, angle, and expression are baked into a single image. The same person can read differently across photos not because the face changed, but because the photo conditions did.
That is why PSL Range is designed for 1–5 photos:
- One photo gives a quick first read.
- Two to four more photos separate the face signal from the photo bias.
- The stable range shows how the face reads across conditions.
- The variance driver points to the dimension that shifts the most — often lighting or angle.
This is not a bug. It is the point. A single number pretends to be exact. A range is honest about uncertainty.
The calibration protocol
Every PSL Range estimate follows the same three-step protocol:
- Quick estimate — start with one clear adult face photo.
- Add photos — add 2 to 4 more photos with varied lighting and angles.
- Read the range — see the stable range, best photo, and main variance driver.
Photos are processed only to generate your result. They are not stored, not shared, and not used for training.
A safer way to read the score
PSL-style scores are AI-generated estimates for photo comparison and entertainment. They are not factual statements about a person, their worth, or their value. The most useful part is not the number alone — it is understanding which photo conditions help or hurt the read.