zkIPV: Zero-Knowledge Proofs for Image Provenance Verification

Abstract

Recent advances in generative AI have significantly improved the ability to create photorealistic synthetic images, including so-called deepfakes, raising concerns about misinformation and the erosion of trust in digital media. Ensuring the integrity and authenticity of images, especially in sensitive domains like journalism, is thus increasingly critical. Existing solutions such as the C2PA (Content Provenance and Authenticity) framework provide origin verification through cameragenerated digital signatures, but fail to account for image modifications that invalidate these signatures. To address this limitation, we propose a zero-knowledge approach to verifiable image editing that preserves both integrity and privacy. This system, ZK-IPV, introduces a practical framework for transforming high-resolution images using zero-knowledge succinct non-interactive arguments of knowledge (zk-SNARKs). ZK-IPV enables developers to specify permissible image transformations, which are then automatically compiled into zk-SNARK circuits. These circuits verify that edits conform to approved operations while concealing the original image content. Furthermore, ZK-IPV supports composable transformations and efficient hashing within proofs, enabling scalable verification pipelines even on commodity hardware. We also formalize the protocol in which an editor can prove that a publicly shared image is derived from a signed original through an authorized transformation, without revealing the original image. This is achieved by demonstrating knowledge of a valid signature and original image such that the verified transformation results in the shared output. Our approach thus extends the C2PA framework with privacypreserving guarantees and post-edit authentication, contributing to the broader goal of trustworthy digital content verification.