How the model reads a face.

You upload one selfie. Our model finds your true skin undertone, eye color, hair value, and natural contrast level. It maps that profile against the 12-season color system used by professional color analysts, picks the season that flatters you most, and returns a full palette plus the shades to avoid. End to end, 15 to 20 minutes.

We run your image through four stages.

• Capture and normalize. We detect 1,024 facial keypoints, isolate your skin, eyes, and hair, and white-balance the photo so room lighting cannot fake an undertone.
• Extract. We pull 30+ color samples from your cheek, mid-face, neck, lip, hair, and iris, then convert them into Lab color space so we can reason about warmth, depth, and saturation independently.
• Classify. A model trained on real human-led color readings scores you across four dimensions, warmth, value, chroma, and contrast, and lands you on one of the 12 seasons.
• Personalize. Your 24-color palette, lipstick and blush picks, wardrobe drapes, and anti-palette are derived from your specific scores, not from a generic season template.

The training corpus is over 3,000 real, human-led color analyses done by certified analysts across the full 12-season system. Every reading in the set has a verified season label and the same structured set of inputs and outputs.

That data is sourced under license, never scraped from social media or customer accounts. Customer selfies are never added to the training set without explicit, separate consent.

Every model release is benchmarked blind against a fresh batch of certified-analyst readings the model has never seen. Our current season-level agreement with a $200+ in-studio reading is roughly 9 out of 10 in normal lighting. The remaining cases tend to be lighting-related, harsh yellow indoor bulbs, AI beauty filters, or a poorly cropped photo, and re-uploading a cleaner selfie almost always resolves them.

We track per-season drift, palette spread, and complaint rate every week. When any of those move in the wrong direction we retrain before shipping new features.

The model does not score you on attractiveness, age, weight, or any other personal feature. It does not detect ethnicity, gender, or identity. It reads color, and only color. Anything else you see in the report, like a styling suggestion or a makeup recommendation, is derived from your color profile, not from any personal feature beyond color.

Your selfie is transmitted over an encrypted channel, processed, and used only to generate your report. We never sell or share it. We do not train future models on it. You can delete your photo and analysis from your account at any time. See the privacy policy for the full picture.

• See a sample report to understand the exact deliverables.
• FAQ for shorter answers to the most common questions.