TokenTrace: Multi-Concept Attribution through Watermarked Token Recovery
Li Zhang, Shruti Agarwal, John Collomosse, Pengtao Xie, Vishal Asnani · 2026
Read TokenTrace: Multi-Concept Attribution through Watermarked Token Recovery by Li Zhang, Shruti Agarwal, John Collomosse, Pengtao Xie, Vishal Asnani (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-19019.txt.
Description
Generative models can copy artistic styles and concepts without credit, and marking often fails when several concepts combine in one image. Li Zhang, with four co-authors, introduces TokenTrace in 2026, a proactive marking frame for robust, multi-concept credit.
Signatures are embedded in the semantic domain by perturbing both the text prompt embedding plus the initial latent noise guiding diffusion generation. For retrieval, a query-based module takes a generated image plus a textual query naming which concepts to retrieve, such as a specific object or style. That mechanism disentangles concepts and verifies each independently from a single image. Experiments show state-of-the-art performance on single-concept and multi-concept credit tasks, beating existing baselines while keeping high visual quality and robustness against common transformations.
Catalog record
| CREATOR | Li Zhang, Shruti Agarwal, John Collomosse, Pengtao Xie, Vishal Asnani |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-19019