← Back to Computer Vision
cs.CV

Can you tell which AI made a 3D model?

Sihan Ma, Siyuan Liang, Dacheng Tao

May 18, 2026

As 3D generative models flood gaming and VR, determining their source matters for authenticity and provenance. This work identifies stable fingerprints that different generators leave behind—cross-view inconsistencies and geometric artifacts—then builds a multi-modal Transformer to detect them across appearance, geometry, and frequency domains. On a new benchmark of 22 generators, the system reaches 97% accuracy with full data and 77% with just five examples per model.
Published as Who Generated This 3D Asset? Learning Source Attribution for Generative 3D Models arXiv:2605.18132
Read the original paper →