Fighting Deepfakes with Zero-Knowledge Proofs

Researchers Daniel Kang, Tatsunori Hashimoto, Ion Stoica and Yi Sun present zk-img, a library that lets image edits be verified with zero-knowledge proofs while keeping the original and intermediate images private. The article says prior ZK-SNARK approaches for images required revealing original images, were limited to images smaller than 128x128, or relied on custom cryptographic arguments or trusted third parties. zk-img is built on the halo2 library and is described as the first library able to privately and securely verify arbitrary transformations on HD images. Benchmarks reported verification times as low as 4.59 ms, and at most 9.32 ms when the output image is also hidden.