zkJPEG: proving a photo was compressed, not faked, without revealing the original

A proof that your JPEG came from a real photo, without showing the photo

A paper posted to the IACR's Cryptology ePrint archive (paper 2026/2039) introduces zkJPEG, a cryptographic tool that proves a published JPEG is the result of correctly compressing an original image that the image owner keeps secret. The work, by Samuel Dittmer, Steve Lu and Kimberlee Model of Stealth Software Technologies together with Joseph Near of the University of Vermont, recently reached the front page of Hacker News.

The problem: legitimate edits break camera signatures

According to the paper, tools for producing photorealistic fakes keep getting easier to use, and one line of defence is camera attestation: a camera signs a photograph at capture time so that viewers can later confirm the image really came from a physical sensor. The catch is that nearly everything people legitimately do to a photo afterwards invalidates that signature. Compressing it into JPEG, blurring a face or redacting a licence plate all change the underlying data, so the signature no longer matches. The authors note that earlier zero-knowledge systems can certify an image's edit history, but they do not survive lossy encoding such as JPEG — the format in which almost every published image actually travels.

What zkJPEG proves

zkJPEG flips that arrangement. The image owner cryptographically commits to the original capture but publishes only the JPEG, accompanied by a zero-knowledge proof that the JPEG is a faithful lossy compression of the committed secret. Anyone can check the proof against the commitment while learning nothing about the hidden input beyond what the published JPEG already reveals.

The paper also folds a broad family of transformations — the authors point to blurring and redaction as examples — into the compression proof itself, so editing and re-encoding can be attested in a single step at what the paper characterises as minimal extra cost.

Built from off-the-shelf parts

Rather than designing a bespoke proof system, the team assembled zkJPEG from existing tooling. They used PicoZK, which turns ordinary Python image-editing code into a zero-knowledge circuit, and ran their proofs on the line-point zero knowledge (LPZK) system. The abstract describes the result as fast and flexible, with the whole construction buildable from widely available ZK components, though it does not include concrete benchmark numbers.

Why it matters

Provenance for images is turning into infrastructure, driven by the same fake-media pressures the paper opens with. zkJPEG points at a workflow that is mostly impossible today: publish a cropped, blurred or redacted JPEG and still cryptographically demonstrate that it descends from a signed camera capture, without ever exposing the unredacted original. That has obvious value for photojournalism, evidence handling and any setting where a raw image contains more than its audience is allowed to see. It also addresses a structural weakness of signature-based schemes, which break the moment an image passes through an ordinary lossy publishing pipeline.

As with any ePrint posting, the construction awaits wider cryptographic review of its security arguments and performance claims, and real-world use would still depend on camera vendors, editing software and standards work. Still, the paper's core suggestion is notable: proving where an image came from and hiding parts of it do not have to be mutually exclusive goals.