Haut.AI released SkinFM, a skin-specific image foundation model trained on more than 4 million skin images and capable of assessing over 230 measurements and 48 visible biomarkers. Alongside it, the company published SkinBench, a public benchmark of 19 tasks across seven common skin concerns, reporting a score of 69.6 for its own model against 50.6 for the highest-scoring general-purpose model, with high false positive rates for eye puffiness and missed deep wrinkles among several large models. Skin remains one of the few aging-relevant surfaces that can be measured non-invasively and repeatedly, making measurement accuracy a precondition for its use as a longevity readout.
Key Points
- SkinFM trained on 4 million images; 230+ measurements, 48 biomarkers
- Specialized model scored 69.6 versus 50.6 for general-purpose leader
- General models showed false positives for puffiness, missed deep wrinkles
Longevity Analysis
Visible skin characteristics — pigmentation, redness, wrinkle depth, periorbital swelling — track underlying processes in barrier defense, microcirculation, fluid balance and tissue repair, which is why skin imaging has drawn interest as a low-friction proxy for biological aging. The value of any such proxy depends entirely on whether the signal is read correctly; a false positive for puffiness or a missed deep wrinkle corrupts the interpretation before any intervention is chosen. Published benchmarks that expose where general-purpose models fail give clinicians a basis for deciding which tools are fit to inform longitudinal tracking rather than one-off assessment.
Original published by LT Wire.