PP-Mark provides provable watermarking for generative AI models
Original title: PP-Mark: Provable and Publicly Verifiable Watermarking for Generative AI
PP-Mark anchors statistical detection to zero-knowledge proofs to ensure unforgeability against forgeries.
Generative AI produces images indistinguishable from real data, but existing watermarking becomes vulnerable to forgeries when detectors are public. PP-Mark combines lightweight statistical detection with zero-knowledge proofs of the embedding process. Forged content cannot pass full verification by construction. It establishes formal unforgeability under cryptographic assumptions and resists black-box imprint forgery, white-box optimization, and regeneration attacks on SD2.1 and SDXL models.
- Resistant to black-box imprint forgery, white-box optimization and regeneration attacks
- Formal unforgeability under standard cryptographic assumptions
- Validated across two latent-diffusion models (SD2.1 and SDXL)
- Practical for decentralized verification
From the original article
1 min read at arxiv.orgGenerative AI models now produce images indistinguishable from real data, making publicly verifiable provenance essential; however, existing watermarking methods become vulnerable to forgeries and adversarial optimization attacks once their detectors are made public.
By Hankyeong Ko · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.