Haut.AI Unveils SkinFM Foundation Model Alongside SkinBench Benchmark Highlighting AI Skin Analysis Flaws
· GlobeWire Daily Editorial Desk · Source: www.prnewswire.com

Image: Haut.AI · Photo source · Resized and letterboxed to 16:9; no AI generation
Haut.AI has introduced SkinFM, an image-based artificial intelligence foundation model trained on over 4 million real skin images to accelerate customized skin measurements. Alongside this release, the company published SkinBench, a public benchmark evaluating six leading AI models from providers such as Google, OpenAI, Anthropic, xAI, and Alibaba.
Key Findings from SkinBench
According to Haut.AI's test results, general-purpose and specialized models frequently struggled with precise dermatological tasks:
- False Positives: Multiple leading models incorrectly identified eye puffiness in 95% to 100% of images where the condition was absent.
- Wrinkle and Type Classification: Models showed difficulties accurately locating wrinkles, such as marking the nose and mouth instead of frown lines, and struggled with consistent skin-type classification.
- Comparative Scores: Haut.AI's specialized model achieved higher balanced accuracy and higher scores on specific tasks like deep forehead wrinkles compared to competing general-purpose models.
- Mixed Results: While Haut.AI scored higher overall on the SkinBench Skin Intelligence Score, certain competing models performed better on specific isolated metrics, such as Gemini 3.5 Flash scoring higher in grading crow's feet and assessing skin tone.
Industry Implications and Verification
The introduction of SkinFM and SkinBench arrives as consumer reliance on generative AI for beauty and skincare recommendations continues to grow. Industry surveys indicate that a significant percentage of beauty consumers are open to or already utilize AI-powered shopping and advice tools.
Experts note that inaccurate AI assessments can lead to mistaken product recommendations, unnecessary spending, and a decline in consumer trust. Beauty brands and researchers can review the full methodology, model comparisons, and the public leaderboard directly at skinbench.ai.
AI-assisted reporting; source claims are attributed and have not been independently verified.



