AI & Acceptable Use Policy
Absolute Commitment — No AI Training on Consumer Data
Consumer Digital Twin data (DTPv3) will NEVER be used to train, fine-tune, distil, or otherwise improve any AI or ML model — internal or external. This is a contractual commitment that supersedes any general AI terms from any sub-processor.
1. Our AI/ML Models
Anproba uses the following AI and ML models within Module B (Digital Twin Pipeline):
- Segment Anything Model (SAM): Body segmentation
- SCHP (Self-Correction for Human Parsing): Clothing and body part parsing
- MoveNet Thunder + OpenPose body_25: Pose estimation
- Stable Diffusion Inpainting + ControlNet: Virtual undressing (output NEVER shown to any user or transmitted to any client)
- PnP SMPL Solver: 3D body mesh fitting
All models are used exclusively for inference on customer data. They are not trained, fine-tuned, or updated using consumer personal data. Model weights are versioned, audited, and updated only from publicly available or properly licensed training datasets.
2. Acceptable Use
The Anproba platform may only be used for:
- Virtual garment try-on for fashion retail and e-commerce
- Clothing size recommendations based on body measurements
- Fit scoring and garment suitability assessment
- Integration into fashion merchant platforms for consumer-facing try-on experiences
3. Prohibited Use
The following uses are strictly prohibited:
- Generating synthetic human images outside of the virtual try-on context
- Facial recognition, identity verification, or biometric identification
- Health inference, medical diagnosis, or body-composition analysis from Digital Twin data
- Age estimation or demographic profiling from biometric data
- Any use of the
bare_bodyDTPv3 field outside the secure processing enclave - Surveillance, tracking, or monitoring of individuals
- Training, fine-tuning, or distilling any AI/ML model using consumer data
- Cross-platform identity linkage without explicit consent
4. Bias and Fairness
Anproba is committed to equitable performance across all body types, skin tones, and demographic groups. We conduct an annual bias audit of Module B outputs across a representative sample of body types (BMI range 15–50+), skin tone diversity (Fitzpatrick Scale types I–VI), and age groups (18–80). Results are reviewed by an independent team and used to improve model performance.
Any systematic bias discovered in production is treated as a P1 incident and addressed within 30 days.
5. Transparency
Anproba discloses to consumers, via the merchant's integration, that: (a) a Virtual Try-On experience is powered by AI; (b) a one-time biometric Digital Twin is created from their body photo; (c) the Digital Twin is encrypted and stored on their device; (d) the original photo is deleted within 30 days. This disclosure is not optional — it is enforced as a requirement of the SDK integration.
6. Human Review
Module B outputs are never used to make automated decisions with legal or similarly significant effects on consumers. Sizing recommendations are advisory, not deterministic. The Virtual Try-On result is a visual aid, not a guarantee of fit.
7. Governance
This AI Policy is reviewed annually and updated to reflect changes in technology, regulation (including the EU AI Act), and industry best practice. Questions: privacy@anproba.de