Gravitate converts five ranks from 1 to 4 into a presentation score: score = 5 × (craft + face + pose + setting + signal), subject to usability caps. With valid ranks, the ordinary range is 25–100. This is a product rubric, not a population attractiveness percentile.
What the five dimensions measure
| Dimension | What we look at | Share |
|---|---|---|
| Craft | Lighting, sharpness and readable framing | 20% |
| Face | Face visibility and natural expression; not facial beauty | 20% |
| Pose | Comfortable body language and framing | 20% |
| Setting | Useful context without distracting clutter | 20% |
| Signal | Visible interests or personal style that give someone a conversation opening | 20% |
How to assign the ranks
The ranks follow the system labels: 1 poor, 2 usable, 3 good, 4 excellent. Apply them separately to each dimension and state an observable reason. These anchors help explain a judgment; they are not a human-validated calibration.
A simple portrait can be excellent for lighting and face visibility without showing a hobby. It does not need top ranks in every dimension to be your lead photo.
From ranks to a score
Each dimension receives an integer rank from 1 to 4. Its dimension score is rank × 25, and its contribution to the total is rank × 5. The ranks are ordinal assessments: the calculation treats them as evenly spaced for a simple, repeatable product score. Equal weights are a design choice, not experimentally established importance.
For example, ranks of 3, 4, 3, 3 and 2 produce 15 + 20 + 15 + 15 + 10 = 75. A score of 75/100 means 75 on this rubric. It does not mean 75% of people find you attractive.
If the subject is absent, the score is capped at 25. A confirmed visual defect caps it at 60. If both apply, the lower cap wins. These checks support usability; they are not a guarantee that every misleading or defective image will be detected.
Where judgment enters the process
The analysis pipeline is designed to ask an AI vision model for the dimension ranks and visible observations. The arithmetic then runs in Gravitate’s own scoring code. Reproducible arithmetic does not make the initial judgments objective; lighting, crops, model updates and ambiguous scenes can change an assessment.
The public example uses deliberately assigned ranks and no uploaded photo. Live photo analysis is currently paused. If an analysis includes an interest or audience suggestion, treat it as a hypothesis based on visible context, not evidence of who will be attracted to you.
What has—and has not—been validated
The scoring logic has automated checks for arithmetic, valid inputs and usability caps. Gravitate has not published a human-rater benchmark, a demographic fairness study or a controlled study connecting this score to dating outcomes. We do not present it as a calibrated measurement of beauty, compatibility or relationship success.
A useful next validation step is to compare specific recommendations with independent human feedback, measure disagreement, and check whether the advice remains useful across age groups, skin tones and photographic conditions. Until that work is available, treat the report as a second opinion.
Turn the score into one useful decision
Read the explanation before comparing numbers. Fix the weakest actionable issue, then decide what role the photo should play. A strong activity photo can stay later in the lineup while a lower-scoring, recognizable portrait goes first. See the worked example.
The photo score is separate from any whole-profile score. Profile text and the number or ordering of images add different considerations; a photo score alone does not evaluate your whole dating profile.
Published by the Gravitate editorial team with AI assistance. Examples are illustrative. Updated . Our editorial approach.