# ZK-IMG: Attested Images via Zero-Knowledge Proofs to Fight Disinformation

**Authors:** Daniel Kang, Tatsunori Hashimoto, Ion Stoica, Yi Sun  
**arXiv:** [2211.04775](https://arxiv.org/abs/2211.04775)  
**Date:** 2022-11-09  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.04775v1/

## Key points

- Recent research proposes using ZK-SNARKs and attested cameras to combat deepfakes
- Existing approaches lack input privacy preservation, are slow, and require custom cryptographic arguments
- The authors introduce zk-img library for attesting image transformations while concealing the pre-transformed image
- zk-img allows developers to specify high-level image transformations compiled into ZK-SNARKs
- Input and output images are hidden by computing their hashes within the ZK-SNARK
- Secure and private methods for chaining multiple image transformations are proposed
- zk-img can securely transform HD images on commodity hardware
- This advancement addresses social disruption caused by deepfakes without compromising user privacy or relying on specialized hardware or cryptographic arguments.

## Summary

To combat the potential disruption caused by deepfakes, recent research has proposed using Zero-Knowledge Succinct Non-interactive Argument of Knowledge (ZK-SNARKs) and attested cameras to verify the authenticity of images taken by a camera. However, existing approaches lack input privacy preservation, are slow (limited to 128x128 images), and often require custom cryptographic arguments. To overcome these limitations, the authors introduce zk-img - a library designed for attesting image transformations while concealing the pre-transformed image. This library allows application developers to specify high-level image transformations which are then compiled into ZK-SNARKs by zk-img. The input and output images are hidden by computing their hashes within the ZK-SNARK. Additionally, this paper proposes secure and private methods for chaining multiple image transformations enabling an arbitrary number of transformations. By combining these optimizations, zk-img is able to securely and privately transform high definition (HD) images on commodity hardware. This advancement is crucial in combating deepfakes and ensuring that image authenticity can be verified without compromising user privacy or relying on specialized hardware or cryptographic arguments. Overall, this paper presents an important contribution in addressing the social disruption caused by deepfakes through the use of zero knowledge proofs and attested cameras. The zk-img library provides a practical solution for verifying image authenticity while maintaining user privacy and efficiency on standard hardware.