Clad Analysis Methodology
Clad structures color and contrast observations from a face photo, estimates a season and subtype, and turns that result into makeup and fashion color guidance. This page describes the current implementation and change-review process without implying an unpublished accuracy score.
Method reviewed: 2026-10-06
What determines my Clad personal color type?
The four seasons correspond to directions on two axes: warm–cool and light–deep. Clad combines diagnostic answers and measured photo signals using defined weights. Warm–cool uses answers and skin-color measurements; the model does not contribute to this axis. Light–deep combines photo measurements and model analysis. If those inputs disagree, the final season can differ from the model’s initial suggestion.
Processing stages
- Choose a gender for your guide or leave it unspecified, then answer two diagnostic questions about sun reaction and skin color. A previous in-person result is optional and can be skipped.
- Check that the format and size of the photo are suitable for analysis.
- Check for a usable face and capture quality, then preprocess the required face region.
- Combine photo observations and diagnostic answers using defined rules to decide the final four-season type.
- Check that the result stays inside the defined ranges and form, and when analysis is unstable, guide a retake or tell you only as much as was confirmed.
- Build palette, makeup, and fashion guidance from the validated type and observations.
A previous in-person result is only for comparison
Entering a previous result does not change this diagnosis, confidence or palette. Gender does not affect the color diagnosis either. If you enter a previous season, the detailed result explains its differences and the recorded evidence for this diagnosis before the survey or satisfaction questions. The comparison distinguishes the same season and differences on warm–cool or light–deep. It does not recreate how your skin responded to earlier physical drapes. Lighting and classification differences are possible factors, not established causes in your case. If you do not remember the previous result, we explain the current evidence. You can continue the diagnosis and view results without entering any previous result.
Confidence and interpretation
Confidence is not the probability that a one-photo result is absolutely correct. It is a supporting signal for how clearly the current model separated the result for this input. Lighting, white balance, filters, color makeup, and camera processing can change the result for the same person.
Evaluation before a change
When we change how analysis works, the change is bound into a single version and compared with the previous one. Evaluation photos and labels are kept in a private set separate from the photos people upload, and identifiers that point to a person are designed to be replaced with values that cannot be turned back. Candidate changes are checked for how results are spread across the seasons, retake decisions, consistency, how long they take, and how often they fail. Result text is checked separately for preserved facts, prohibited claims, language that does not match the reader, and the wording shown when analysis cannot finish. Changes to written content go through a review that hides which version is which.
Clad does not publish a single promotional accuracy percentage until a representative, independently reviewable sample and its labeling basis can be published together.
What the analysis does not infer
- The identity of the person in the photo
- Race, health status, disease, or a skin-condition diagnosis
- Personality, ability, or social traits
- A definitive judgment that replaces an in-person drape consultation
A reproducible self-check
Take two or three front-facing photos in the same place with the same lighting and camera settings, then check whether the direction repeats. If results vary substantially or confidence is low, retake in natural light and consider an in-person drape session for an important decision.