ZK-WAGON: Imperceptible Watermark for Image Generation Models using ZK-SNARKs

View on arXiv ← Back to list

Authors: Aadarsh Anantha Ramakrishnan, Shubham Agarwal, Selvanayagam S, Kunwar Singh

Published: 2025-10-02 12:39:57+00:00

Comment: Accepted at AI-ML Systems 2025, Bangalore, India, https://www.aimlsystems.org/2025/

AI Summary

ZK-WAGON introduces a novel system for imperceptibly watermarking image generation models using Zero-Knowledge Succinct Non-Interactive Arguments of Knowledge (ZK-SNARKs). This approach enables verifiable proof of origin for AI-generated images without exposing sensitive model internals or degrading image quality. It employs Selective Layer ZK-Circuit Creation and LSB steganography to securely embed and verify cryptographic proofs in images from both GAN and Diffusion models.

Abstract

As image generation models grow increasingly powerful and accessible, concerns around authenticity, ownership, and misuse of synthetic media have become critical. The ability to generate lifelike images indistinguishable from real ones introduces risks such as misinformation, deepfakes, and intellectual property violations. Traditional watermarking methods either degrade image quality, are easily removed, or require access to confidential model internals - making them unsuitable for secure and scalable deployment. We are the first to introduce ZK-WAGON, a novel system for watermarking image generation models using the Zero-Knowledge Succinct Non Interactive Argument of Knowledge (ZK-SNARKs). Our approach enables verifiable proof of origin without exposing model weights, generation prompts, or any sensitive internal information. We propose Selective Layer ZK-Circuit Creation (SL-ZKCC), a method to selectively convert key layers of an image generation model into a circuit, reducing proof generation time significantly. Generated ZK-SNARK proofs are imperceptibly embedded into a generated image via Least Significant Bit (LSB) steganography. We demonstrate this system on both GAN and Diffusion models, providing a secure, model-agnostic pipeline for trustworthy AI image generation.

Key findings
The system successfully demonstrates imperceptible watermarking of AI-generated images from both GAN and Diffusion models, providing verifiable proof of origin without visual distortion. It ensures model-agnostic and hardware-efficient verification by leveraging ZK-SNARKs and selective circuit creation (SL-ZKCC), making the process secure and tamper-evident. The approach allows third-party users to verify image authenticity without exposing proprietary model information or private inputs.
Approach
ZK-WAGON creates ZK-SNARK proofs of image origin by selectively converting key layers of an image generation model into a ZK-SNARK circuit (SL-ZKCC) to reduce computation. These proofs, compressed and signed, are then imperceptibly embedded into the generated image using Least Significant Bit (LSB) steganography. During verification, the proof is extracted from the image, decompressed, and checked against the original model via a public verification key, confirming the image's authenticity and origin.
Datasets
UNKNOWN
Model(s)
TinyGAN, BigGAN, Stable Diffusion 2.1 Base, Tiny AutoEncoder for Stable Diffusion (TAESD), EZKL toolkit, Halo2 proving system, Lilith
Author countries
India
← Previous
Next →