Artworks AI enforces a strict service policy to prevent the upload and processing of images featuring minors.
Our AI department has developed a proprietary solution based on Vision-Language Model (VLM) technology to process and describe each uploaded image.
Technology
Ther VLM analyes each uploade image and assings ratings (from 0-100) in 3 categories: baby, child, adolescent
Next the VLM analyzes each uploaded image and assigns a rating (from 0-100) in the category adult.
From the first 3 scores an overall “minor” score is derived:minor = baby + child + adolescent
If minor > 90% all processing is declined, the user notified with a warning. The image is transferred for further analysises and later deleted.
Test results:
Kids and adolescents
baby: 98.367child: 1.2282adolescent: 0.0959adult: 0.3088
baby: 25.71
child: 68.2316
adolescent: 2.4273
adult: 3.6309
baby: 1.0365child: 36.6997adolescent: 54.8043adult: 7.4594
baby: 4.1343
child: 93.8656
adolescent: 0.623
adult: 1.377Adults
baby: 14.5602
child: 2.488
adolescent: 13.3212
adult: 69.6306
baby: 30.6425
child: 12.1200
adolescent: 2.1897
adult: 55.0479
baby: 19.2775child: 1.3424adolescent: 7.4535adult: 71.9266
baby: 0.2274 child: 2.5965 adolescent: 66.3567 adult: 30.8193Manual validation
Additionally, every image check is stored in a database for manual validation to assess the efficiency of the algorithm and facilitate future fine-tuning.
You can test our algorithm efficiency live at the undress page.
If you find any issues or have suggestions for improvement, please write us at [email protected]
