CryptoWorld News reported that XBIT DEX said that a developer, Alosh Denny, has cracked Google’s AI watermark detection system, which has embedded invisible watermarks in over 100 billion AI-generated contents. Using only 200 sample images and some mathematical methods, Denny reverse-engineered the system and built “reverse-synthid,” an open-source tool designed to remove watermarks until detection fails. While removing nearly 91% of the watermarks’ spectral features, it achieved a 16% evasion rate with almost no visible quality loss. The project has been open-sourced and has received more than 4,100 stars on GitHub.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 4
  • 1
  • Share
Comment
Add a comment
Add a comment
GateUser-8df0eb2b
· 5h ago
91% removal rate + almost no loss, want to delve into the technical details
View OriginalReply0
L2ArbitrageTrader
· 5h ago
Open source fighting against big corporations' black boxes, this is very Web3
View OriginalReply0
On-ChainSoilAfterTheRain
· 5h ago
reverse-synthid this name sounds quite provocative
View OriginalReply0
LatencyMonk
· 5h ago
Just 200 samples are enough for reverse engineering, indicating that the watermark design itself has structural flaws.
View OriginalReply0
  • Pinned