ZK-JPEG: Zero-Knowledge Image Editing and Compression
gslin
71 points
14 comments
September 19, 2026
Related Discussions
Found 5 related stories in 75.3ms across 7,105 title embeddings via pgvector HNSW
- Regressive JPEGs vitaut · 15 pts · July 18, 2026 · 56% similar
- Knowledge Compressor Mappleton · 11 pts · August 25, 2026 · 53% similar
- The case against JPEG XL contact9879 · 91 pts · September 14, 2026 · 52% similar
- Proof of Capture: Apple Reference Image, but open source and using steganography merybenavente · 88 pts · September 10, 2026 · 46% similar
- DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression mfiguiere · 94 pts · September 17, 2026 · 45% similar
Discussion Highlights (5 comments)
Retr0id
Interesting work. > our tool can verify a large family of image transformations Is this family large enough to transform real image A into arbitrary fake image B? > ZK-JPEG can merge transparent or translucent layers into an image, useful for placing visual watermarks, creating double exposures, or merging visual layers, potentially AI generated over portions of the image. In our tool, the transparent layer is revealed, but anything beneath an opaque portion of the layer becomes secret. Note that in the case of an AI generated layer, the layer itself is revealed, minimizing the dishonesty in using generative AI (but not minimizing the dishonesty of stealing artwork to train the AI) Seems like the answer to my question is "yes", so you have to carefully inspect the transformations to see if they look legit. (The parenthetical was a surprising inclusion in an academic context)
jamienk
Hasn't this crossed over into being a philosophical question? You want to attest to reality or provenance. But then you start making lists of "acceptable" changes: file format; compression; color-correction; size; taking "medium" into account... It then slowly slides into "average human perceptibility"; "irrelevant details"; keeping the "spirit" of the image intact; judging the intention or motive of the user or of the viewer; aligning the image with "values"; appropriate/legal use... etc. Not sure what the end-game is? We are trying to make images in the AI era be as reliable as they were pre-AI? But they were not reliable pre-AI, it was just more difficult to intentionally doctor them.
mvid
Tie this into the Apple/Android/Sony/Leica signed photos, and you get provenance from capture to publish
AmazingEveryDay
I'm trying to think of recent real-world examples where the reality of the photo was of major importance. Photographs just don't seem to have the same significance, even as potential verification of their realness becomes more achievable.
pixelsort
If this ever became widespread, people would game the "verified real" checkmark using the analog gap. 1. Print AI photo 2. Point fancy expensive camera at printed photo 3. Take a "real" photo Then you'd see more sensors and even more expensive cameras. Then you'd see miniatured virtual-production volumes. So, this is not a solution. It's just an arms race disguised as one.